You are on page 1of 66

Quantum Optics Theory

Dr. Michael J. Hartmann 2010/2011

Contents
1 1.1 1.2 1.3 1.4 Quantisation of the Electromagnetic Field . . . . . . . . . . . . Fock or Number States . . . . . . . . . . . . . . . . . . . . . . Coherent States . . . . . . . . . . . . . . . . . . . . . . . . . . Squeezed states . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Single-mode squeezed states . . . . . . . . . . . . . . . 1.4.2 Multi-mode squeezed states . . . . . . . . . . . . . . . Quantum correlations and the Einstein-Podolsky-Rosen paradox . . . . . . . 3 4 6 6 8 8 9 10 11 12 12 12 13 13 14 15 16 16 16 17

1.5 2 2.1

2.2 2.3 2.4 3 3.1

Coherence properties of the electromagnetic eld . . . . . . . . . 2.1.1 Field correlations . . . . . . . . . . . . . . . . . . . . . . 2.1.2 Optical coherence: Youngs double slit experiment . . . . First order optical coherence . . . . . . . . . . . . . . . . . . . . Second order optical coherence: Photon correlation measurements Phase dependent correlations: Homodyne detection . . . . . . . .

Representations of the Electromagnetic Field 3.1.1 Expansion in number states . . . . . 3.1.2 Expansion in coherent states . . . . . 3.1.3 The Wigner function . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

4 4.1

18 Open quantum systems . . . . . . . . . . . . . . . . . . . . . . . 19 4.1.1 Damped Harmonic Oscillator . . . . . . . . . . . . . . . 21 4.1.2 P-Representation and Fokker-Planck Equations . . . . . . 23 25 Interactions Between Radiation and Atoms . . . . . . . . . . . . 26 5.1.1 Long-Wavelength Approx. and Dipole Representation . . 26 1

5 5.1

5.1.2 5.1.3 5.1.4 5.1.5 5.1.6 5.1.7

Two-Level Atoms and the Jaynes-Cummings Model . . . Spontaneous Emission of a Two Level Atom . . . . . . . Resonance Fluorescence . . . . . . . . . . . . . . . . . . Power Spectrum of the Emitted Light . . . . . . . . . . . Equations for Correlation Functions and the Quantum Regression Theorem . . . . . . . . . . . . . . . . . . . . . . Raman Transitions and Electromagnetically Induced Transparency . . . . . . . . . . . . . . . . . . . . . . . . . . .

27 28 28 30 32 34 36 37 38 39 39 41 45 46 46 46 48 49 50 51 52 54 56 58 59 60 60 62 62 63 64

6 6.1 Cavity Quantum Electrodynamics 6.1.1 Strong driving regime . . 6.1.2 Strong coupling regime . . 6.1.3 Input-output formalism . . 6.1.4 Circuit QED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7 7.1 Light Forces on Atoms . . . . . . . . . . . . . . . . . . . . 7.1.1 Concept of Doppler cooling . . . . . . . . . . . . . 7.1.2 Semiclassical theory of light forces . . . . . . . . . 7.1.3 Standing wave Doppler cooling . . . . . . . . . . . 7.1.4 Limit of Doppler cooling . . . . . . . . . . . . . . . 7.1.5 Cooling beyond the Doppler limit: Sisyphus cooling 7.1.6 Optical lattices . . . . . . . . . . . . . . . . . . . . 7.1.7 Many-particle representation: 2nd quantisation . . . 7.1.8 Interactions between ultra-cold atoms . . . . . . . . 7.1.9 Mott insulator to superuid quantum phase transition 7.1.10 Measuring the Mott insulator and superuid phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

8 8.1 Trapped Ions . . . . . . . . . . . . 8.1.1 Trapping potential: Paul trap 8.1.2 Manipulations with lasers . 8.1.3 Sideband cooling . . . . . . 8.1.4 Trapping multiple ions . . . 8.1.5 Ion trap quantum computer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 1

1.1

Quantisation of the Electromagnetic Field


B = 0 E = E = 0 B = 0 0 E t B t (1.1) (1.2) (1.3) (1.4)

The electromagnetic eld in vacuum is described by Maxwells equations

The electric eld E and the magnetid eld B can be expressed in terms of a vector potential A via E = A and B = A, where A obeys the Coulomb gauge t condition A = 0 (1.5) and the wave equation A = 1 2A , c2 2t (1.6)

where c is the vacuum speed of light. Maxwells equations (1.1-1.4) follow from eqs. (1.5) and (1.6) so that one only needs to solve (1.6) under condition (1.5) to nd the electromagnetic eld. This can be done by expanding the eld A in terms of eigenfunctions of the Laplacian . h ak uk (r)eik t + c.c. , (1.7) A(r,t) = 20 k
k

where c.c. is for complex conjugate of the preceding term and the uk (r) satisfy the gauge condition uk (r) = 0 and the eigenvalue equation uk =
k

2
k 2 c

uk .

(1.8)

Here 2 = c2 k k and the uk (r) form an orthonormal and complete set of eigenh functions, hence a basis. The prefactor 20 in equation (1.7) is chosen such k that the ak are dimensionless. For a quantisation volume of a cube with edge length L and periodic boundary conditions, the uk are plane waves,

1 uk = e eikr , L3 where the direction of e is the polarisation of the eld. 4

(1.9)

The Hamilton function of the eld is, H= 1 2 d3r ( A) ( A) + , 0 0 (1.10)

where the canonically conjugate momentum to A is = 0 A. This form of the Hamiltonian can be justied by showing that the Euler-Lagrange equations for the corresponding Lagrangian reproduce Maxwells equations, (1.1) - (1.4). Using eqs. (1.7) and (1.9), H can be written as a sum of harmonic oscillators, p2 H =
k k

2
k

q2 , k

(1.11)

with, qk (t) = pk (t) = i h a eik t + c.c. 2k k (1.12)

hk ak eik t c.c. . (1.13) 2 qk and pk play the roles of positions and momenta since their equations of motion, qk = pk and pk = k 2qk (1.14)

also follow from the Hamilton equations for H as in (1.10). This also shows that (1.10) is indeed the Hamiltonian for the eld. In analogy to the harmonic oscillator, the electromagnetic eld is now quantised by promoting qk and pk to become operators qk and pk and imposing, qk , pk = i k,k . h The eld now reads, A(r,t) =
k k

(1.15)

h ak uk (r)eik t + c.c. , 20 k

(1.16)

where a and ak are creation and annihilation operators for the quanta in mode k, ak , ak = 0 and ak , a = k,k .
k

(1.17)

These are bosonic commutation relations and show that the quanta of the eld, the photons, are bosons. The operator A(r,t) is here written in Heisenberg picture d i and fulls dt A = h A, H with H as in (1.10). 5

1.2

Fock or Number States

The quantised electromagnetic eld reads, A(r,t) =


k

h ak uk (r)eik t + c.c. , 20 k

(1.18)

where a and ak create and annihilate photons in mode k, i.e. ak , ak = 0 and


k

ak , a k

= k,k . The operator a ak is therefore a number operator and its eigen


k

values nk denote the number of photons in a mode k. a ak |nk = nk |nk .


k k k

(1.19)

The eigenstates |nk of a ak are called Fock or number states, where a ak |0 = 0 denotes the vacuum state without photons. Since H = k hk (a ak + 1 ), the 2 k vacuum energy diverges. This has however no measurable consequences since only energy differences can be observed. Since they are eigenstates of an Hermitian operator, number states form a complete orthonormal basis, nk |n p = k,p (1.20) (1.21)

|nk
nk

nk | = 1 .

Number states are useful for calculations but difcult to generate experimentally.

1.3

Coherent States
D() = ea

A coherent state is generated by applying the shift operator


a

(1.22)

onto the vacuum, | = D()|0 (1.23) Such a shift is for example generated when a classical oscillating dipole emits an electromagnetic eld. Here, the dipole-eld interaction reads, V = ed(t) E(t) i ( a a ), h 6 (1.24)

where we have omitted fastly oscillating terms in the approximation. The time evolution of this interaction Hamiltonian generates a shift D() with = t. The shift operator D() has the properties D () a D() = a + D () a D() = a + D () = D1 () = D() D() = e Coherent states have the following properties: eigenstates of a a| = | expansion in Fock states | = e
||2 2 ||2 2

(1.25) (1.26) (1.27)

ea e a .

(1.28)

(1.29)

n |n n=0 n!

(1.30)

contain an indenite number of photons normalised | = 1 probability to measure n photons is Poisson distributed ||2n ||2 e P(n) = n! not orthogonal | = e| | overcomplete d 2 | | = (1.34)
2

(1.31)

(1.32)

(1.33)

1 the position distribution for a harmonic oscillator, H = h(a a + 2 ), that is initially in a coherent state | = | ||ei is a Gaussian with constant vari h ance x = 2m and a mean value x(t) = 2|| cos(t ) that oscillates with frequency .

1 [x x(t)]2 P(x) = (1.35) exp 2 2x 2x In this sense, a coherent state of a harmonic oscillator is the state which comes closest to our classical (i.e. non-quantum) notion of a harmonic oscillator. 7

1.4

Squeezed states

The expectation value and variance of an operator A in some state | are dened as, A = V (A) = (A)2 = A = |A| AA
2

(1.36) (1.37)

The uncertainty relation for two operators A and B reads, A B 1 | [A, B] | . 2 (1.38)

For the two eld quadratures X1 = a + a and X2 i(a a ), the uncertainty relation reads, X1 X2 1. (1.39) States, for which X1 X2 = 1 are called minimum uncertainty states.

1.4.1

Single-mode squeezed states

States, for which either X1 < 1 ( and consequently X2 > 1) or X2 < 1 ( and consequently X1 > 1) are called squeezed states. Squeezed states can for example be generated in a degenerate parametric amplier via processes where a drive of frequency 2 generates two photons of frequency at a time. The Hamiltonian for this process reads, HI = i h a2 (a )2 , 2 (1.40)

in an interaction picture with respect to the photon energy H0 = ha a. Writing = ||e2i we nd for the quadratures Y1 = aei + a ei and Y2 = i(aei a ei ), V (Y1 (t)) = e2||t V (Y1 (0)) and V (Y2 (t)) = e2||t V (Y2 (0)), (1.41)

i.e. the variance in Y1 becomes squeezed and the variance in Y2 becomes amplied. We thus dene the squeezing operator to be, S() = exp and a squeezed state to be, |, = D()S()|0 . 8 (1.43) 2 2 a (a ) , 2 2 (1.42)

Writing = re2i with r = ||, we nd S ()aS() = a cosh(r) a e2i sinh(r) S ()a S() = a cosh(r) ae2i sinh(r). (1.44) (1.45)

For a harmonic oscillator, H = h(a a + 1 ), that is initially in a squeezed state 2


h with |, = |||, re2i , we nd for the position operator q = 2 (a + a ) that its mean value oscillates harmonically just as for a coherent state,

q(t) =

2 h || cos(t).

(1.46)

In contrast to a coherent state the variance (q)2 is however not constant but oscillates with frequency 2, (q)2 (t) = h cosh2 (r) + sinh2 (r) 2 cosh(r) sinh(r) cos(2t 2 ) . (1.47) 2

1.4.2

Multi-mode squeezed states

Multi-mode squeezed states can for example be generated in a non-degenerate parametric amplier via processes where a drive of frequency 1 + 2 generates one photon of frequency 1 and one photon of frequency 2 at a time. The Hamiltonian for this process reads, HI = i a1 a2 a a , h 1 2 (1.48)

in an interaction picture with respect to the photon energies. We thus dene the two-mode squeezing operator to be, S(G) = exp G a1 a2 Ga a , 1 2 (1.49)

and a two-mode squeezed state to be, |1 , 2 = D(1 )D(2 )S(G)|0 . The expansion of two-mode squeezed vacuum in terms of number states reads, | = S(G)|0, 0 =
1 tanhn(r)|n, n cosh(r) n=0

(1.50)

This state is entangled, that means there is no way of writing, | = |1 |2 (1.51)

As a consequence, if we measure the number of photons in mode 1 to be n for the state | of eq. (1.50) we know with certainty that mode 2 has n photons as well although the variance in photon number is not zero for | . Entanglement is the key resource for the exponential speed-up of quantum computers as compared to classical computers. 9

1.5

Quantum correlations and the Einstein-PodolskyRosen paradox


X j = a j ei + a ei j

For a two mode squeezed state consider the quadratures ( j = 1, 2), with Xj ,Xj
+ 2

= 2i ,

(1.52)

which are canonically conjugate. Correlations between both modes can be quantied by 1 2 X1 X2 V ( , ) = (1.53) 2 For the two mode squeezing dynamics generated by the Hamiltonian (1.48) we have V ( , )(t) 0 for t , so that X1 and X2 become perfectly correlated. For such a situation, Einstein, Podolsky and Rosen argued as follows: Suppose the photons of both modes are in sufciently separated laboratories so that no communication at the speed of light can take place within the runtime of the measurements. Assume further that V ( , ) = 0. If now X1 is measured, X2 is known with certainty. Therefore mode 2 must have been in an eigenstate of X2 , X2 | = | before the measurement. Since modes 1 and 2 are so far apart that they can not interact, mode 2 must have been prepared in state | before the photons have been taken apart. If however, the experimentator in the laboratory of mode 1, after the photons have been separated, decides to measure X1 with = , he or she would nd that mode 2 must have been in the eigenstate X | = | before the photons have
2

been separated. Since there are however values for and , for which X2 , X2 = 0 and consequently | = | , this reasoning would require that mode 2 has been prepared in two non-equal state at the same time. The resolution to the paradox is that there are states with non-local correlations. Information can however still only propagate at the speed of light or slower, since both experimentators can not nd out that their measurement results are actually correlated without exchanging information. The existence of such non-local correlations has been veried several times in experiments. For example Ou and Mandel measured the variances V (X1 ) and +/2 V (X2 ) for a two mode squeezed state. Provided V ( , )(t) 0, one has V (X1 ) V (X2 ). With this reasoning, Ou and Mandel found a product of the +/2 variances of V (X2 )V (X2 ) = 1.4 0.02, whereas the uncertainty relation +/2 would require V (X2 )V (X2 ) 4 if one measured V (X2 ) directly. 10

Chapter 2

11

2.1
2.1.1

Coherence properties of the electromagnetic eld


Field correlations

If a photon is detected and hence absorbed by a photon detector it disappears from the eld. A eld in the state (density matrix) thus changes according to E (+) E () in the detection process. Since the information about the nal state | f of the eld after the detection is irrelevant, the measured photon intensity reads, I(r,t) = f |E (+) E () | f = Tr E () E (+) (2.1)
|f

More generally, one can dene the eld correlations, Gn (x1 , . . . , xn , xn+1 , . . . , x2n ) = Tr E () (x1 ) E () (xn )E (+) (xn+1 ) E (+) (x2n ) (2.2) where xn = (rn ,tn ). These correlations have the properties, G(1) (x, x) 0 Gn (x1 , . . . , xn , xn , . . . , x1 ) 0 G(1) (x1 , x1 )G(1) (x2 , x2 ) (a a1 )2 (a a2 )2 1 2 G(1) (x1 , x2 ) a a1 a a2 1 2
2

(2.3) (2.4) (2.5)


2

(2.6)

2.1.2

Optical coherence: Youngs double slit experiment

The electric eld impinging at r on a screen which emerges from two pin holes at positions r1 and r2 is a superposition of the eld E1 from pin hole 1 and the eld E2 from pin hole 2, E (+) (r,t) = E1 (r,t) + E2 (r,t)
(+) (+)

(2.7)

Pin holes with sizes smaller than the wavelength can be treated as point sources that emit spherical waves. The eld E (+) (r,t) thus reads, E (+) (r,t) i h eit a1 eiks1 + a2 eiks2 , 8L0 R (2.8)

where s1 = |r r1 |, s2 = |r r2 | and we have approximated s1 s1 R1 . The 1 2 intensity on the screen thus shows interference fringes with maxima for k(s1 s2 ) + = 2n (n integer), i.e. I(r,t) a a1 + a a2 + 2| a a2 | cos[k(s1 s2 ) + ] . 1 2 1 (2.9)

12

2.2

First order optical coherence


G(1) (x1 , x2 ) G(1) (x1 , x1 )G(1) (x2 , x2 )

The g(1) -function, which is dened by, g(1) (x1 , x2 ) = , (2.10)

obeys |g(1) (x1 , x2 )| 1 due to eq. (2.5). The visibility of the interference fringes, dened by, Imax Imin = . (2.11) Imax Imin
I1 can be expressed in terms of the g(1) -function via the relation, = |g(1) (x1 , x2 )| 2I1 +II2 , 2 where I1 = G(1) (x1 , x1 ) and I2 = G(1) (x2 , x2 ).

2.3

Second order optical coherence: Photon correlation measurements

The probability to detect a photon at time t and a second at time t + is given by the 2nd order correlation function G(2) () = E () (t)E () (t +)E (+) (t +)E (+) (t) , which is in fact independent of t for a steady beam. In analogy with rst order optical coherence, we dene the g(2) -function, g(2) () = G(2) () G(1) (0)
2

(2.12)

For an innitely large time difference , the two photon detection events are expected to become independent, g(2) () 1. For 0 on the other hand, we distinguish two scenarios, g(2) (0) > 1 called bunching and g(2) (0) < 1 called anti-bunching. Before proceeding to discuss several states of the eld, we note that, V (n) n , (2.13) g(2) (0) = 1 + n2 where n = a a and V (n) = Examples: a a
2

n2 .

coherent states, E (+) | = E (+) | : G(2) () = E () (t)E () (t + )E (+) (t + )E (+) (t) = G(1) (0) g(2) () = 1. 13

(2.14)

uctuating classical eld G(2) () = d 2 E P(E )E (t)E (t +)E (t +)E (t). g(2) (0) = 1 + d 2 E P(E ) |E |2 |E |2 |E |2
2 2

(2.15)

Hence, whenever P(E ) 0 we have bunching, g(2) (0) < 1. This means that every eld that can be generated by mixing classical (coherent) elds will display bunching. Consequently, if a eld displays anti-bunching, it is quantum in the sense that it cannot be generated by mixing classical elds. a number state |n shows anti-bunching as follwos form eq. (2.13). g(2) (0) = 1 1 n (2.16)

squeezed states, |, with = re2i , can show bunching if they are phase squeezed, = and real. Amplitude squeezed states, = 0 and 2 real, can show anti-bunching for ||2 2 sinh2 (r) cosh2 (r). The squeezed vacuum, = 0, shows bunching.

2.4 Phase dependent correlations: Homodyne detection


Consider a beam splitter with transmittivity that combines elds with annihilation operators a and b to an output eld c via the relation, c= a + i 1 b. (2.17)

If the eld with annihilation operator b is in a coherent state with large amplitude , the measured number of output photons obeys, c c (1 )| |2 + | | (1 ) X+ , 2 (2.18)

where X+ = aei + a ei and = | |ei . In eq. (2.18) we have ne2 glected a term 2 a a since | | 1. After subtracting (1 )| |2 , the c thus allows to measure the quadrature X photon number c + . Chang2 ing the phase of the reference eld b by /2 then allows to also measure the canonically conjugate quadrature X .

14

Chapter 3

15

3.1
3.1.1

Representations of the Electromagnetic Field


Expansion in number states

Any density state of the electromagnetic eld can be expanded in number states, =

n,m=0

cn,m |n m|.

(3.1)

This particularly convenient for states which are diagonal in this basis, such as thermal states, = 1e where
kh T B

n=0

kh n T B

|n n| =

n 1 1+n 1 + n n=0

|n n|

(3.2)

n = Tr(a a)

is the expectation value for the photon number.

3.1.2

Expansion in coherent states

Since the coherent states form an over-complete set, any state can as well be expanded in terms of coherent states. Observing that the matrix elements n|O|m , are the expansion coefcients of the Taylor series 2 ( )n ()m (3.3) e|| |O| = n!m! n|O|m n!m! n,m=0 for any Hermitian operator O, we see that all matrix elements n|O|m can be determined from only the diagonal elements |O| . Therefore the P-function, dened by = d 2 P() | | (3.4)

contains all information about the state and is thus an alternative form to represent it. For a coherent state we obviously have P() = (2) ( 0 ). P() is not always positive and can therefore in general not be interpreted as a probability distribution. This is evident from g(2) (0) = 1 + d 2 P() ||2 ||2 ||2
2 2

(3.5)

where ||2 = d 2 P() ||2 . For anti-bunching, P() must have negative or highly singular parts. The same holds for squeezing with either V (X1 ) < 1 or V (X2 ) < 1 (X1 = a + a , X2 = i(a a ) since, V (X1,2 ) = 1 + d 2 P() [ ( )]2 16 (3.6)

Whenever P() is not a probability distribution, the corresponding eld cannot be generated by mixing classical elds and is thus regarded as quantum. Antibunching and squeezing are thus signatures of quantum elds.

3.1.3

The Wigner function


() = Tr ea

The characteristic function is dened as


a

(3.7)

and the Wigner function is its Fourier transform, W () = 1 2 d 2 e

() =

1 2

d 2 Tr e(a

) (a)

. (3.8) and

Splitting the variables into real and imaginary parts, = + i, a = =


q+ip 2 h

q+ip 2 h

we get, W (q, p) = 2 h (2)2


d d Tr ei(qq)i(pp) ,

(3.9)

where = h2 and = 2 , and see that W (q, p) is a Fourier transform of h . It therefore contains all the information about the state . The Wigner function is a Gaussian convolution of the P-function, W () = 2 d 2 P( ) e2| |
2

(3.10)

For a coherent state |0 , the Wigner function is Gaussian with variances 1, centred at 0 , 2 1 2 2 (3.11) W (x1 , x2 ) = e 2 (x1 +x2 ) , where x1 = 2Re[ 0 ] and x2 = 2Im[ 0 ]. For a squeezed state, on the other hand, the variance is squeezed in one direction and enhanced in the conjugate one, W (x1 , x2 ) =
2 x1 2 exp 2r 2e

exp

2 x2 2e2r

(3.12)

with real ( = 0) and positive squeezing parameter = r > 0. Finally, a number state has the Wigner function, W (x1 , x2 ) = 2 (1)n Ln (4r2 )e2r , (3.13)

2 2 with r2 = x1 + x2 , which can become negative. This is another signature that number states are non-classical.

17

Chapter 4

18

4.1

Open quantum systems

In no realistic situation, quantum systems of interest are perfectly isolated. Hence an interaction between the system under study and its environment has to be taken into account. Since the environment is typically extremely large, it is impossible to solve its dynamics exactly and approximations are needed. These approximations make use of the fact that the interaction between system and environment is typically weak. In the following we will derive an equation of motion for the reduced density matrix of the system under study and then simplify it via two approximations to obtain an ordinary rst order differential equation. The Hamiltonian of the total system can be written as a sum of the Hamiltonian for the system (HS ), the environment (HE ) and interaction (HI ), H = HS + HE + HI (4.1)

If we write the density matrix of the whole system R in a product basis of system states |s and environment states |e , R = , , , c, ; , |s , e s , e |, the reduced density matrix for the system reads, S = TrE (R) =

, ,

c, ; , |s s |

(4.2)

The dynamics of R is given by the Liouville equation, R = i[H, R] (4.3)

where we have set h = 1 and H is as in eq. (4.1). Since eq. (4.3) is linear in R, we can write as a shorthand R = iLR i[H, R] and R(t) = exp(iLt)R(0) exp(iHt)R(0) exp(iHt). Analogously to the Hamiltonian, we may write L = LS + LE + LI . We now split the density matrix R into a relevant part in which we are interested, Rrel = PR, and an irrelevant part, Rirr = QR. Here, P and Q are projection operators that full P 2 = P, Q 2 = Q and 1 = P + Q. For our purposes we will use for P a projector whose action on any operator O is given by, PO = E TrE (O). (4.4) Using these projection operators and the identity, ei(L1 +L2 )(tt0 ) = eiL1 (tt0 ) i
t t0

ds eiL1 (ts) L2 ei(L1 +L2 )(st0 ) ,

(4.5)

which is valid for any two operators L1 and L2 , we get the equations, P R(t) = iPLPR(t) iPLQR(t) = iPLPR(t) iPLQeil(tt0 ) R(t0 ) (4.6) 19

and P R(t) = iPLPR(t)


t t0

ds PLQeiLQ(ts) LPR(s)iPLQeiLQ(tt0 ) R(t0 ) (4.7)

To continue, we now make three assumptions, 1. the environment state E is invariant under the environment dynamics, [HE , E ] = 0 2. the initial state factorises, R(t0 ) = S (t0 ) E 3. the system environment interaction fulls PLI P = 0. Observing that, PLE = LE P = 0 and [P, LS ] = 0, and using these three assumptions, we get the Nakajima-Zwanzig equation for the relevant part of the density matrix, t Rrel (t) = iLS Rrel (t) ds PLI QeiLQ(ts) LI Rrel (s) (4.8)
t0

This is an exact equation for the degrees of freedom of the system only. It however contains a time integral over the history Rrel (s < t) and is thus not Markovian. To obtain a Markovian, ordinary differential equation for S , we now make the following two approximations, 1. weak coupling approximation: we only keep the leading order terms in the coupling LI in eq. (4.8). Higher orders would approximately contain additional powers of HI (t t0 ). This approximation should thus be good as long as ||HI ||(t t0 ) 1. 2. Markov approximation: It will turn out that the kernel of the integral decays fastly. For the time range that signicantly contributes to the integral in eq. (4.8), we can thus approximate S (t) ei(HS +HE )(ts) S (s)ei(HS +HE )(ts) . This means that for the time range that signicantly contributes to the integral, S (s) in the integral kernel at an earlier time s can be linked to S (t) without taking HI into account. Using these approximations, we get the equation,
t

S (t) = i[HS , S (t)]

t0

ds TrE {[HI , [HI (s t), E s (t)]]} ,

(4.9)

where we have switched back to the Hamiltonian picture and HI (s t) = ei(HS +HE )(ts) HI ei(HS +HE )(ts) . Eq. (4.9) is now a Markovian ordinary differential equation since the right hand side only depends on s (t). This equation is a good approximation if the assumptions 13 are met and if the integral kernel decays much faster than s (t) can change due to the system-environment interaction. Fortunately this scenario is often true for very large environments. 20

4.1.1

Damped Harmonic Oscillator

We will now apply the master eq. (4.9) to a specic model. To this end we assume a harmonic oscillator that couples to an environment formed by harmonic oscillators, H = HS + HE + HI with, HS = 0 a a + HI = 1 , 2 HE = k b bk +
k k

1 2

(4.10) (4.11)

gk (a + a)(b + bk ) k
k

and

E =

1 HE exp Z kB T

For evaluating the master eq. (4.9), we need HI in the interaction picture, HI (st). This can be simplied further via the so called Rotating Wave Approximation, HI (s t) a (s t)ei0 (st) + a (s t)ei0 (st) , (4.12)

where (s t) = k gk bk eik (st) . This approximation is good since the terms that are neglected in eq. (4.12) belong to transitions that would not conserve the energy of the system. Therefore their effect on the evolution of the system is strongly suppressed. Expanding the double commutator in the rhs of eq. (4.9), we nd four terms, Tr {[HI , [HI (s t), E S (t)]]} = Tr {HI HI (s t)E S (t)} Tr {HI (s t)E S (t)HI }Tr {HI E S (t)HI (s t)}+Tr {E S (t)HI (s t)HI } . As an example we will explicitly calculate the rst one, Tr {HI HI (s t)E S (t)} = (s t) aaei0 (st) + (s t) aa ei0 (st) + (s t) a aei0 (st) + (s t) a a ei0 (st) S (t) For evaluating the rhs of eq. (4.9) we thus need to calculate the integrals,
t

I1 = I3 =

ds (s t) ei0 (st) , ds (s t) ei0 (st) ,

t0 t t0

I2 = I4 =

ds (s t) ei0 (st)

t0 t t0

ds (s t) ei0 (st)

Due to the form of E , see eq. (4.10) and the coupling, see eq. (4.12), the rst and fourth integral vanishes, i.e. I1 = I4 = 0. For I2 we nd,
t

I2 =

t0

ds gk g p Tr bk b E ei(0 p )(st) = p
k,p

0 tt0

d |gk |2 nk ei(0 k )
k

21

Since the environment is large, we can introduce a density of states per frequency interval, (), and write,
0

I2 =

d
tt0 0

d ()|g()|2 n()ei(0 ) 2 ei0 d (i0 )+1 tt0


0

If we now assume that ()|g()|2 n() = c for some positive , we nd,


0

d
tt0 0

d i(0 ) +1 e = 0 ( + 1) 2

We see that the kernel in the integral decays as (i0 )(+1) and rapidly goes to 1 zero for 0 . This asymptotics emerges since the bath correlations decay rapidly. We can thus extend the lower bound of the time integral to and nd,
0

Re

( + 1)ei0 = (0 )|g(0 )|2 n(0 ). (i)+1

Here the imaginary part only leads to a redenition of the frequency 0 . If one already knows that the bath correlations decay rapidly, a faster route of calculation can be made by extending the time integration to the range (, 0) right away and using the relation,
0

dtei(0 )t = ( 0 ) iPV

1 , 0

(4.13)

where PV denotes the Cauchy principal value. With this calculation one nds, (4.14) I2 = n(0 ) + i and I3 = [n(0 ) + 1] + i, 2 2 where = (0 )|g(0 )|2 ,
d ()|g()|2 2 0 n().

= PV

d ()|g()|2 2 0

[n() + 1]

and

= PV and are merely frequency shifts and we will assumed they have been absorbed into 0 . With these results we can write down the master equation, S = i[0 a a, S ] + n(0 )(2a S a aa S S aa ) (4.15) 2 + [n(0 ) + 1](2aS a a aS S a a) 2 From eq. (4.15) we nd the amplitude of the oscillator is damped, a (t) = a (0)e 2 t , and that a a (t) = a a (0)et + n(0 )(1 et ). One can also show that the oscillator will, for t , be damped to a thermal state with the same temperature as the environment,
t

lim S =

1 HS exp Z kB T 22

(4.16)

4.1.2

P-Representation and Fokker-Planck Equations

For solving the master equation (4.15), various representations for the density matrix S can be used. One version is to represent S in the number basis, = n,m n,m |n m| (I will skip the index S from now on). Here we will use the representation in terms of the P-function, = d 2 P() | |. For representing equation (4.15) in a basis of coherent states | , we dene, || = e 2 || | .
1 2

(4.17)

Using equation (1.29) and the expansion in Fock states, see eq. (1.30), we nd, ||a = || a || = || a || = || || ||a =

(4.18)

After an integration by parts, that uses P() 0 for || , we get the following correspondences for the translation of equation (4.15) into an equation for P(), a P a P a a P P a P a a P a

(4.19)

We move to an interaction picture with respect to H0 = 0 a a + 1 and get with 2 the help of the above mappings an equation of motion for the P-function in interaction picture, PI , PI (,t) = t 2 + + n(0 ) 2 PI (,t). (4.20)

This form of equation is known as a Fokker-Planck equation and also occurs in classical stochastic processes. The analogy to classical stochastic processes becomes even more apparent with the following notation in terms of the real (x1 ) and imaginary (x2 ) parts of , = x1 + ix2 . x= x1 , x2 =
x1 x2

A = x, 2

D = n(0 )

10 , 01

(4.21)

23

for which the Fokker-Planck equation (4.20) takes the form, PI (x,t) = t A + 1 2 2 D j,l PI (x,t). j,l=1 x j xl
2

(4.22)

Here A is called drift vector and D diffusion matrix since the dynamics of d the mean values x j (t) = dx1 dx2 x j PI (x,t) only depends on A via dt x j = A j and the dynamics of the variances V (x j ) = x2 x j 2 depends on A and D, i.e. j d V (x j ) = V (x j ) + n(0 ). We thus see that mean values decay at a rate /2 dt and variances approach their assymptotic value V (x j ,t ) n(0 ) at a rate , x (t) = x (0)e 2 t ,

V (x j ,t) = V (x j , 0)et + n(0 ) 1 et .

(4.23)

Therefore, any squeezing that might be present in the initial state is damped away at a rate . Steady state solutions Often it is sufcient to know the asymptotic steady state of a master equation which fulls t P = 0. From eq. (4.22) we see that this implies,
xj

A j + 1 2 x D j,l P = 0 and try the ansatz, 2 l=1


l

1 A jP = D j,l P 2 l=1 xl

2 D j,l D j,l xl ln(P) = 2A j xl . l=1 l=1

(4.24)

Dening a function via P(x) = N 1 exp( (x)), where N is a normalisation, we see that plays the role of a potential for a generalised force F via the relation F = , where 2 2 D l,k Fj = 2 D1 Al . (4.25) j,l l=1 k=1 xk Here we have assumed that the matrix D is invertible. The existence of the potential is then guaranteed if the force F fulls the conditions j Fl = l Fj . As a consequence can be expressed via an integral of F along a path in the complex plane which is independent of the path taken,
x

(x) =
x1 s1 ds1 0 n(0 ) x2 s2 ds2 0 n(0 )

F d
2 2 x1 +x2 2n(0 ) ,

(4.26)

where d is the line element along the path. For our example given in equation (4.21), we nd (x) = + = and therefore, (4.27)
2 x2 + x2 1 exp 1 2n(0 ) 2n(0 )

P(x) =

This is a Gaussian distribution centred at x = 0 with variance n(0 ) in all directions. 24

Chapter 5

25

5.1

Interactions Between Radiation and Atoms

Since electromagnetic elds couple to charges and atoms are composed of a positively charged nucleus and negatively charged electrons, electromagnetic elds couple to atoms. As the nucleus is much heavier than the electrons, only the latter will be moved by the coupling to the elds. For one electron with charge e, the Hamiltonian describing this interaction reads, H= 1 p + eA(r) 2m
2

eV (|r|) + Hrad ,

(5.1)

where m is the mass, p the momentum and r the position of the electron. A and V are the vector potential of the electromagnetic eld and the Coulomb potential respectively. Hrad is the energy of the free eld. Using the Hamilton equations of motion for the classical Hamilton function corresponding to (5.1), one can see that these reproduce the known form of the Lorenz force, mr = eE er B. The form of the coupling in Hamitonian (5.1) can furthermore be motivated by observing that the probability to nd a charged particle at a location r, |(r,t)|2 is invariant under a local change of the phase of the wave function , (r,t) (r,t) = (r,t)ei (r,t) . Of course this transformation would not leave the dynamics of the charged particle invariant. Therefore this change in the dynamics should be caused by a coupling to elds which transform accordingly. One can show that the Schr dinger equation corresponding to the Hamiltonian (5.1), i t = H, o h is invariant under the transformation, (r,t) (r,t)ei (r,t) , A A + , V V , t (5.2)

with = h (r,t), which leaves both invariant, |(r,t)|2 as well as the physical e elds E and B. This formally justies the form of the coupling.

5.1.1

Long-Wavelength Approx. and Dipole Representation


1 p j + eA(r j ) 2m j
2

We consider a system of charged particles centered at r = 0 with Hamiltonian, H =


j

+
j,l

q j ql 1 + Hrad , 80 |r j rl |

(5.3)

where the second term describes the Coulomb interaction of the particles with charges q j . For atoms we have |r j rl | aBohr 104 with the Bohr radius aBohr . We thus assume that the particles are distributed over a volume much smaller than the wavelength of the radiation, |r j rl | , so that we can approximate A(r j ) A(0) and neglect any coupling of spins to the magnetic eld. 26

The electric properties of the particle are then described by their dipole moment d = j q j r j and we can write the Hamiltonian (5.3) in a new form by api plying the unitary transformation T = exp( h d A(0)). We nd T r j T = r j , h T p j T = p j + q j A(0) and Tak, T = ak, + (iuk, d)/ 2 k 0 , where uk, is the mode function of the mode with vector k and polarization . The transformed Hamiltonian reads, H = T HT = Hatom + Hrad d E(0),
p2 q q

(5.4)

j l 1 where Hatom = j 2mj j + k, |uk, d|2 + j,l 80 |r j r | . l For the relevant case of alkali atoms, only the outer valence electron is moved by the radiation elds. We thus represent the Hamiltonian (5.4) in its eigenstates, Hatom | = E | and nd for the coupling term,

d E(0) = i h
hk 20V

, ,k

g, ,k ak g, ,k a | |,
k

(5.5)

where g, ,k =

e |d|

5.1.2

Two-Level Atoms and the Jaynes-Cummings Model

1 Since |E| V , atom photon interactions are strongest for light that is conned to a small volume around the atom. This is achieved in cavities, like pairs of mirrors hat face each other, and causes the spectrum of the light to be discrete (only wavelength that vanish at the mirrors are possible). The description of the atom-light interactions can thus be restricted to only one atomic transition | | = |e g| and one mode ak . Since h|g, ,k | Ee Eg , we apply a rotating wave approxi mation and arrive at the Jaynes-Cummings Hamiltonian,

hA z + hC a a + h(g a + g + a), (5.6) 2 where z = |e e| |g g|, = |g e| and + = |e g|. Since HJC conserves the number of excitations, [HJC , |e e|+a a] = 0, we diagonalise it for each excitation number n separately and nd for A = C the eigenvalues and eingenstates, HJC = |n 1, e |n, g (5.7) 2 For each excitation number n, we thus nd a doublet of states |n, + and |n, , separated by an energy 2 |g|2 n. If the atom is initially excited and no photons h are present, the excitation oscillates back and forth between atom and radiation mode at a frequency 2 n|g|. These oscillations are called Rabi oscillations and 2 n|g| the Rabi frequency. En, = hC n 1 h 2 |g|2 n and |n, = 27

5.1.3

Spontaneous Emission of a Two Level Atom

Any atom always interacts with its electromagnetic environment according to the Hamiltonian (5.4), i.e. H= h0 z + hk a ak, i (gk, a gk, + ak, ), h k, k, 2
k, k, h

(5.8)

k h0 , where counts the polarisations and gk, = 20V e d. Since |gk, | the modes ak, can be treated as a quantum environment in a master equation approach. For 0 1015 Hz, the thermal photon number at room temperature is n0 1013 and we can assume that all ak, are in their vacuum states. The dynamics of the atomic levels is thus described by the master equation,

= i[

0 z , ] + ( + + + ), 2 2
3 |d|2

(5.9)

The spontaneous emission rate = 30h c3 can here be calculated with the same 0 techniques as the damping rates in equation (4.15). Using the commutation relations of the j and the cyclic property of the trace, we nd the dynamics, |e e| (t) = (t) = |e e| (0) et (0) e
(i0 +/2)t

(5.10) (5.11)

The occupation probability of the excited state thus decays at rate and coherences at rate /2.

5.1.4

Resonance Fluorescence

We now consider an atom that is continuously driven by a laser eld and emits photons into the surrounding vacuum. This phenomenon is called resonance uorescence. The driving laser corresponds to a coherent state for one mode aL with the free time evolution, |l (t) = eiL aL aL t |l ,

where l (t) = eiLt L (0)

(5.12)

For solving the full problem, one would need to solve the dynamics generated by the Hamiltonian (5.8) for the initial state |, 0 = |L |0k=kL |atom , 0 . One can show that this dynamics is equivalent to the dynamics generated by the Hamiltonian, H = H i L (t)+ + i L (t) h h (5.13) 28

for an initial state |, 0 = |0k |atom , 0 where all modes ak are in vacuum. In equation (5.13), H is given by eq. (5.8) and L (t) = gL L (t). For calculating the dynamics of the atomic degrees of freedom, we switch to an interaction picture with respect to H0 = hL z and write down the master 2 equation corresponding to H, where we treat all radiation modes as a quantum environment, = i[ z iL + + iL , ] + ( + + + ), 2 2 (5.14)

where = 0 L and L = gL L . Note that now all radiation modes are in vacuum, hence the simple form of eq. (5.14). Assuming L to be real, we can now calculate the equations of motion for the expectation values + , and z . These can be written in compact form, v = Av + b, where v = ( + , , z )T , b = (0, 0, )T and i /2 0 iL 0 i /2 iL A= 2iL 2iL (5.15)

(5.16)

Noting that the solution to the differential equation f(t) = f (t) + is f (t) = cet /, where c is an integration constant, we see that limt f (t) = /, provided that Re[] < 0. For equation (5.15) this means, that a well dened asymptotic state exists provided that all eigenvalues of A have negative real parts. This is the case for A and also implies that A1 exists. The steady state solution can therefore be found from vss = A1 b and reads, + z
ss ss ss

= i =

2( + i)L 2 + 42 + 82 L +
ss

(5.17) (5.18) (5.19)

2 + 42 2 + 42 + 82 L

From these equations one sees that z ss < 0, which means that a simple laser drive can not generate a population inversion and the occupation probability of |g is always higher than for |e .

29

A time dependent solution of eq. (5.15) can be found by observing and AvSS = 0 from d (v vSS ) = A (v vSS ) . dt We nd z (t) = 1 + +

d dt vSS

=0

(5.20)

3 82 3 1 e 4 t cosh(t) + sinh(t) (5.21) 2 + 82 4 3 2i 3 (t) = 2 1 e 4 t cosh(t) + + sinh(t) , 2 + 8 16 2

where = 1 4 162 . For || > /8, is imaginary and the solutions oscil2 late while approaching vSS .

5.1.5

Power Spectrum of the Emitted Light


h2 k k uk uk eik t eik (t+) a ak 2 k 40 d

The correlation function E () (r,t)E (+) (r,t + ) =

k,k

d 2 hn ei 3 (2)3 20 c

(5.22)

can be written as a Fourier transform of the emitted energy per frequency interval as the second line of equation (5.22) shows. Here the d-integration is over the entire solid angle. For deriving (5.22) we consider the steady state regime where E () (r,t)E (+) (r,t + ) is independent of t and hence a ak = k,k nk for k > k 0 and zero otherwise. The power spectrum, S(r, ), can thus be written as the Fourier (back) transform of E () (r,t)E (+) (r,t + ) , S(r, ) = lim 1 t 2

dei E () (r,t)E (+) (r,t + )

(5.23)

Since the master equation (5.14) only predicts the dynamics of the atomic degrees of freedom but equation (5.23) species the power spectrum in terms of the emitted eld, we need to nd the relation between the emitted eld and the atomic degrees of freedom.

30

Relation between emitted eld and state of atom From the Hamiltonian (5.13) one can derive the Heisenberg picture equations of motion, ak = ik ak gk h g
k

(5.24) h ak + z (t) (5.25)

= i0 + z
k

We can formally integrate equation (5.24) to obtain, ak (t) = eik t ak (0) gk h


t 0

ds (s)eik (ts) .

(5.26)

Plugging this into the expression for E (+) , and integrating over the entire solid angle of k yields, E (+) (r,t) = d 8 2 0 c2 r
0

d 2 ei c ei c
0

ds (s)ei(ts) . (5.27)
2 iy 0 d e

The integral over d can now be calculated as follows: 2 2! 2 iy = iy3 , where > 0. We get, 0 d e (iy)3
0

E (+) (r,t) =

d 8 2 0 c2 r

t 0

ds (s)

1
r st c

1
r st + c 3

(5.28)

The ds-integration can be done via the residual theorem. To this end we note r 3 r 3 0 for s < 0 and that s t c 0 for that for t large enough, s t c |s| . In the complex plane we can thus integrate along the whole real line and close the contour along the half-circle |s| . Applying the residual theorem yields, 2 0 d r r (+) E (r,t) = (t + ) (t ) , (5.29) 2r 40 c c c where we have approximated
d2 2 (s) 0 (t r/c) since 0 ds2 s=tr/c r + c ) can be ignored as it represents an incoming

||, |gk |. The term with (t eld. If one takes into account the polarisation of the radiation eld, one obtains the general result, E (+) (r,t) =
2 0 d r r r d (t ), 2r 40 c r r c

(5.30)

31

5.1.6

Equations for Correlation Functions and the Quantum Regression Theorem

Quantum Regression Theorem If a set of operators Y j full the equations of motion d Y j (t) = G j,l (t) Yl (t) , dt l then the following equations of motion for correlation functions hold, d Yk (t)Y j (t + ) = G j,l () Yk (t)Yl (t + ) . d l (5.32) (5.31)

For proving the quantum regression theorem, one shows that Yk (t)Y j (t + ) can be read as an expectation value, Yk (t)Y j (t + ) = Tr(Y j ()), where () = eiH (t)Yk eiH and Y j (t) = Tr(Y j (t)). Therefore equations (5.32) hold with the corresponding initial conditions. We now apply (5.32) to equations (5.20) to nd for the vector + (t)+ (t + ) + (t)+ (t) u = lim + (t) (t + ) + (t) (t) t + (t)z (t + ) + (t)z (t) the equations d u = Au. d (5.34)

(5.33)

Using + (t)+ (t) = 0, + (t) (t) = 1 (1+ z (t) ) and + (t)z (t) = z (t) ), 2 we nd from these for G() = + (t) (t + ) , 42 2 ei0 e(i0 + 2 ) X+ e(i0 + 4 ) X e(i0 + 4 +) + + G() = 2 , + 82 2 + 82 2 2 2 (5.35) where X =
2 34 2 +82 4 2 4 and = 8 1 2 2 4
3 3

162 .

Power spectrum Using equation (5.30), the power spectrum can now be calculated as S(r, ) = I0 (r) 2

d ei G() =

I0 (r) Re

d ei G() ,

(5.36)

32

with I0 (r) = 40 c2 r d r r . The second equality in equation (5.36) folr r 0 lows from the observation that G() = G () together with a substitution of integration variables. The second expression in equation (5.36) contains integrals of the form

2d

L() = Re
0

d ei(0 ) =

, ( 0 )2 2

(5.37)

with , 0 real and > 0. The function L() is called a Lorenzian. It has a peak at = 0 , L(0 ) = 1/, and its width at half the maximum is , L(0 ) = 1/(2). We thus see that the rate of exponential decay of a correlation function is equal to the with of a spectral line as given by the Lorenzian in equation (5.37). After performing the integration in equation (5.36), we nd for a weak laser drive, || /8, 42 ( 0 ) (5.38) S(r, ) = I0 (r) 2 + 82 and for a strong laser drive, || /8, I0 (r) 42 1 2 S(r, ) = 2 2 ( 0 ) + 2 2 + 8 2 ( 0 )2 + ( 2 )2
3 3

(5.39)

1 1 4 4 + + 2 + ( 3 )2 4 [ (0 + 2)] 4 ( (0 2))2 + ( 3 )2
4 4

In the weak driving limit, the spectrum shows one delta peak at = 0 , whereas in the strong driving limit it shows one delta peak at = 0 and three Lorenz peaks at = 0 and = 0 2. This is the famous Mollow spectrum. Photon statistics Apart from the spectral properties of the emitted light, one can also look at its photon statistics. Of particular interest here is the probability to detect two photons at a specic time separation at the detector. This is quantied by the g(2) -function as dened in equation (2.12). Here we nd, g(2) () = 1 e 4 cosh() +
2 3

3 sinh() , 4

(5.40)

with = 1 4 162 . Importantly, we have anti-bunching, g(2) (0) = 0, indi2 cating that the photons are emitted one at a time and never two or more together. This is due to the fact that the atom can only take one excitation at a time and after one photon emission the atom needs to be reexcited before a second photon can be emitted. 33

5.1.7

Raman Transitions and Electromagnetically Induced Transparency

Here we make use of the fact that only specic transitions between atomic levels have a non-vanishing dipole moment, j|d| j . More specically, j|d| j = 0 only if l j = l j 1 and m j = m j or m j 1, where L2 | = hl(l + 1)| and Lz | = hm| . We thus choose an atom the couples to two classical laser elds according to the Hamiltonian H = 3 33 + 2 22 + (1 eiL1t 31 + 2 eiL2t 32 + H.c.) (5.41)

where we have chosen the atomic levels such that the transition |1 |2 has no dipole moment. For example the 87 Rb D2 -line levels 52 S1/2 (F = 1) for |1 , 52 S1/2 (F = 2) for |2 and 52 P3/2 (F = 1) for |3 . This implies that any coupling to electromagnetic elds on that transition is weaker by a factor aBohr / as compared to the dipole transitions. Therefore spontaneous emission losses from level |2 are negligible. In an interaction picture with respect to H0 = L1 33 +(L1 L2 )22 the Hamiltonian becomes time independent and reads, H = 33 + 22 + (1 31 + 2 32 + H.c.) with = 3 L1 and = 2 (L1 L2 ). Raman Transitions We now choose the lasers to be in a two photon resonance, = 0 and go to another interaction picture with respect to H0 = 33 , HI (t) = 1 eit 31 + 2 eit 32 + H.c. and formally integrate the Schr dinger equation o i |,t + T |,t = T T
t+T t d dt |,t t+T

(5.42)

(5.43)

= iHI (t)|,t to get


s

dsHI (s)|,t

1 T

ds
t t

drHI (s)HI (r)|,t +. . . (5.44)

in an iterative expansion. We now consider a scenario where, |1,2 | and |1,2 |2 T 1 T (5.45)

In this regime, we nd that the dominant contributions to the right hand side of equation (5.44) come from the second term and that all other contributions are negligible. Due to
|1,2 |2 T

1 we can write 34

|,t+T |,t T

d dt |,t

and can thus

d approximate (5.44) by an effective Schr dinger equation dt |,t = iHI,e f f |,t o with 1 2 |1 |2 + |2 |2 |1 |2 |2 |2 HI,e f f = 33 11 22 12 + H.c. , (5.46) Level |3 thus decouples and two photon transitions |1 |2 appear.

Electromagnetically Induced Transparency For two photon resonance, = 0, the Hamiltonian (5.42) has an eigenstate |0 = 2 |1 1 |2 |1 |2 + |2 |2 (5.47)

which has no component in level |3 and hence does not contribute to spontaneous emission. Therefore it is called a dark state. If the eld 1 is much weaker than the eld 2 , the dark state approaches |1 . We now study the propagation of a weak eld E1 = E1 (z)eiL1t + c.c. where 31 1 = dh E1 in dependence of the strong eld 2 . We consider a one-dimensional setting, where the wave equation in a medium reads, 1 2 2 2 2 2 E = 0 2 P = 2 2 E (5.48) c2 t 2 z t c t where the susceptibility is dened by P = 0 E and the polarizability in our case reads, P = natoms d13 31 eiL1t + d23 32 eiL2t + c.c. (5.49) Here, natoms is the density of the atoms and the density matrix of one atom in the interaction picture with H0 . To calculate we solve the maser equation for one atom that is driven by the two lasers and decays via spontaneous emission from level |3 at a rate and from level |2 at a much smaller rate . In doing so we make a perturbative expansion in the weak probe eld E1 . To linear order we nd for the susceptibility , = 2(2 i) natoms |d13 |2 . h0 (2 i)(2 i) 82 2 (5.50)

We see that 0 for = 0 and 0. In this case the eld E1 decouples from the polarization of the medium as can be seen from equation (5.48) and the medium becomes transparent, even for = 0. Furthermore the group velocity of 2 1 dn , where n = 1 + , can E1 becomes vg = c(n L1 d )1 c 1 + natoms |d13 | 2 2 h
0

become much smaller than c for light.

natoms |d13 |2 2 0 2 h 2

1. This phenomenon is called slow

35

Chapter 6

36

6.1

Cavity Quantum Electrodynamics

Since the interaction strength between light and atoms g scales as g 1/ V where V is the quantisation volume, it can be enhanced by conning the light eld in a small volume, a cavity. In the classical example of a Fabry-P rot cavity formed by e two mirrors that face each other, the electric eld has to vanish at the mirror surfaces due to their high reectivity. Hence the possible wavelength for resonance modes for a cavity of length L are n = 2L , n (6.1)

where n is a positive integer. As the spectrum is discrete, an atom typically only couples with one transition to one single cavity mode. The Hamiltonian that describes this system reads, H = 0 |e e| + c a a + g(a|e g| + H.c.) + (eiLt a + H.c.), (6.2)

where 0 is the transition frequency of the atom, c the resonance frequency of the cavity mode and represents a drive of the cavity by a classical laser eld. The photon annihilation operator for the cavity mode is a and |e (|g ) denote the excited (ground) state of the atom. Since the atom-cavity system is not perfectly isolated, interactions with the environment (here the electromagnetic vacuum) need to be taken into account. This can be done with a master equation that describes the two dominant loss processes, spontaneous emission of the atom and the leakage of photons out of the cavity. It reads, = i[H, ]+ (2 + + + ) 2 + (2aa a a a a), 2

(6.3)

where is the density matrix of cavity mode and atom, = |g e|, + = |e g| and and are the rates of sponatneous emission respectively cavity decay. As the Hamitonian (6.2) contains an explicit time dependence of the laser eld, it is convenient to work in an interaction picture with respect to H0 = L |e e| + L a a. In this frame, the Hamiltonian can be taken to read, H= z + a a + g(a|e g| + H.c.) + (a + H.c.), 2 (6.4)

where z = |e e| |g g|, = 0 L and = c L . The daming terms in equation (6.3) in turn stay invariant. 37

6.1.1

Strong driving regime

From equation (6.3) with Hamiltonian (6.4) one can derive the following equations of motions for operator expectation values, d a = i a i ig dt 2 d = i a + ig az dt 2 d z = ( z + 1) 2ig a+ a dt

(6.5) ,

which are nonlinear and not closed. Whereas they are difcult to solve for a general case one can nd an approximate solution for the regime of a strong input drive, i.e. || |g|. To this end we write a = + a, where is the coherent part of the eld, = a , and a the quantum uctuations around it. For strong input drive one can expect that ||2 a a and neglect the uctuations a. This turns the equations (6.5) into a closed set and one nds for the steady state d d d with dt a = dt = dt z = 0, n z ss = 1 + n0 (1 + 2 )
1

(6.6)

where n = ||2 is the number of photons in the coherent part of the cavity eld, 2 n0 = 2g2 and = 2 . One nds, z ss 1 for n n0 and z ss 0 for n n0 . In the former case the atom is in its ground state |g and in the latter 1 in the fully mixed state 2 (|e e| + |g g|). The quantity n0 sets the number of photons needed to excited the atom and is thus called critical photon number. Moreover z ss < 0 so that no population inversion can be achieved. For a laser resonant with the cavity, = 0, one furthermore nds for the photon number n in the steady state, ||2 [n + (1 + 2 )n0 ]2 , 2 (6.7) g2 where C = 2 is called the cooperativity. Since equation (6.7) is a 3rd order polynomial in n it has 3 solutions. These are not always physically relevant since the photon number n can only be positive. There exists however the possibility of having 3 steady states which in fact happens for = 0 and sufciently strong drive . Which of the steady states will be reached then depends on the initial conditions for the system. 0 n3 + 2(1 8C + 2 )n0 n2 + (1 + 2 )[(1 + 8C)2 + 2 ]n2 n = 4

38

6.1.2

Strong coupling regime

We now focus on a regime of strong coupling, that is g and g , and assume a weak driving laser, i.e. || |g|. The dynamics of the system is described by eq. (6.3). Ignoring the weak drive and dissipation for now, we can approximate the Hamiltonian by the Jaynes-Cummings model, c.f. eq. (5.6). Since HJC conserves the number of excitations, [HJC , |e e| + a a] = 0, it can be diagonalised for each excitation number n separately. For C = A , the eigenvalues and eigenstates are given in eq. (5.7). Photon blockade A test whether a cavity indeed works in the strong coupling regime can be done via an effect called photon blockade. This effect is due to the nonlinearity of the spectrum, that is the fact that E2, 2E1, . We now focus > on C = A and nd from (5.7) that E2, 2E1, = (2 2)|g| > 0. A laser that is resonant with the transition |0 |1, is thus not resonant with the transition |1, |2, . Provided || |g| it can thus only bring one excitation into the cavity and occupy state |1, but it can not bring a second excitation into the cavity, that is the cavity blocks a second photon from entering it. The effect can be veried by measuring the photon statistics for the light leaving the cavity. In this output light, the photons only come one by one but never in pairs or bunches. The output light is thus anti-bunched as can be quantied by its g(2) -function, c.f. eq. (2.12). The output eld outside the cavity is related to the light eld in the cavity via the input-output formalism which we discuss next.

6.1.3

Input-output formalism

We now consider a Fabry-P rot cavity for which light can only be emitted to one e side, say the positive x-direction. An electric eld outside the cavity then reads (ignoring the polarisation) E = E (+) + E () with E (+) = i
k

hk bk eik (tx/c) 20V

(6.8)

where [bk , b ] = k,k . For highly reective cavity mirrors, photons inside the k cavity will only couple to modes outside with k C and we can approximate
h = 20C . Note that k in the exponential cannot be approximated since V this would cause substantial errors for large t. We can thus dene a new eld outside the cavity, hk 20V

1 b(x,t) = eiC (tx/c) 2 39

d b()ei(tx/c)

(6.9)

where = C . With this denition, b (x,t)b(x,t) is the number of photons that pass point x per time unit, i.e. the photon ux at x. The coupling of photons in the cavity to modes outside can be described by,

V=

d h() a b() + ab ()

(6.10)

where h() is the coupling strength. Hence the Heisenberg equations of motion for a and b() read, b(,t) = ib(,t) ih()a(t) (6.11)

a(t) = i[HJC , a(t)] i

d h()b(,t)

(6.12)

By formally integrating eq. (6.11), b(,t) can be specied in terms of its initial condition at t0 < t, b(,t) = ei(tt0 ) b(,t0 ) ih() tt0 dsei(ts) a(s), or its nal condition at t1 > t, b(,t) = ei(tt1 ) b(,t1 ) + ih() tt1 dsei(ts) a(s). These expressions can then be plugged into eq. (6.12). Since the coupling h() is weak (|h()| C ), the cavity eld only couples to a narrow frequency range of elds outside and we can approximately take h() to be constant in this range. We thus dene h2 () = 2 , where the meaning of is the rate at which photons leak out of the cavity. With this approximation we nd, a(t) = i[HJC , a(t)] a(t) + ain (t) (6.13) 2 a(t) = i[HJC , a(t)] + a(t) aout (t) (6.14) 2
i i where ain = 2 d ei(tt0 ) b(,t0 ) and aout = 2 d ei(tt1 ) b(,t1 ) are the input and output elds. From (6.13) and (6.14) we nd the input-output relation, aout (t) = a(t) + ain (t), (6.15)

which is valid for any intra cavity Hamiltonian, not only for HJC . From eqs. (6.13-6.15) and causality, it can be shown that for any intra cavity operator X, X(t), aout (t ) = (t t) X(t), a(t ) (6.16) where () = 1 for > 0, () = 1/2 for = 0 and () = 0 for < 0. From these commutation relations one can show for cases where ain (t)| = 0 (Note that the laser does not contribute here.) that, a (t)a (t)aout (t)aout (t) = 2 a (t)a (t)a(t)a(t) out out a (t)aout (t) out = a (t)a(t) (6.17) (6.18) X(t), ain (t ) = (t t ) X(t), a(t ) ,

Hence anti-bunching of photons in the cavity results in photon anti-bunching in the output eld. The photon blockade effect is thus visible in the output eld. 40

6.1.4

Circuit QED

Since charges and currents generate electromagnetic elds and interact with them, electronic circuits can be used to store elds. To keep oscillating elds for as long as possible, the resistance of the circuit should be minimal and superconducting circuits are thus the best candidates for this task. Basic elements of such circuits are a capacitances and inductances. The energy stored in a capacitance is Ecap = Q2 2C where C = Q U (6.19)

is the capacitance, Q the charge and U the voltage drop. The energy stored in an inductance in turn is 2 Eind = where L = (6.20) 2L I is the inductance, the magnetic ux and I the current. The magnetic ux is linked to a corresponding voltage drop by Faradays law of induction, U= d dt (6.21)

The two basic elements can be combined in a row to form an LC-cicuit, which is described by the Hamiltonian, HLC = Q2 2 C + = 2C 2L 2 d dt
2

2 , 2L

(6.22)

and is an electronic form of a harmonic oscillator with mass mLC = C and fre h quency LC = 1/ LC. Hence it is quantised by imposing [Q, ] = i . Transmission line We now discuss the quantisation of electromagnetic elds carried by a transmission line. A transmission line can be modelled by a sequence of LC-circuits with the Hamiltonian, C Htl = j=1 2
N

d j dt

1 ( j j1 )2 , j=2 2L

(6.23)

and is thus an electronic form of a harmonic chain. The continuum limit of this model can be taken by dening a capacitance per unit length, c = C/d and an inductance per unit length, l = L/d, where d is the distance between neighbouring

41

capacitances respectively inductances. Observing that, ( j j1 )/d / z Ltl /2 and N d L /2 dz, where Ltl is the length of the transmission line, we nd, j=1
tl

Htl = 2 0

c dz 2 Ltl /2
h e

Ltl /2

d (z) dt

1 + 2l

(z) z

(6.24)

where = /0 and 0 =

is the ux quantum (e is the elementary charge). Ltl /2 The corresponding Lagrange function L = 2 L /2 dzL with Lagrange den0 tl sity L = c 2 1 (z )2 leads to the Euler-Lagrange equations, L d L
2 2l L z (z ) dt

= 0, which in our case read, 2 d2 v2 2 = 0. dt 2 z (6.25)

This is a wave equation that describes electromagnetic waves propagating in z direction at velocity v = 1/ lc. Transmission line resonator A resonator or a cavity for the light elds in a transmission line can be engineered by implementing boundary conditions for the electromagnetic elds which make their spectrum discrete. This is simply done by cutting the transmission line a two points such that two capacitances emerge. The boundary conditions at these two = = 0 since no currents points, Ltl /2 and Ltl /2 are then z z
z=Ltl /2 z=Ltl /2

can ow across the capacitances. The solution of equation (6.25) then reads, (z,t) = 2 (2n + 1) 2n z n,e(t) cos Ltl z + n,o(t) sin Ltl n=0 Ltl , (6.26)

and the Hamiltonian (6.24) decomposes into a sum of harmonic oscillators for c 2 c 2 c 2 c 2 2 2 each mode, Htl = 2 2 n,e + 2 n,e n,e + 2 n,o + 2 n,o n,o , where n,e = 0 n=0 (2n+1)v 2nv and v = 1/ lc. Hence, the eld can be quantised by quantisLtl , n,o = Ltl ing each of the harmonic oscillators separately, just as for a free electromagnetic eld. We choose the length of the resonator such that only the mode 2, e is of interest to us. We thus nd a ux eld (z,t) an the corresponding voltage eld U(z,t), (z,t) = 0 U(z,t) = i h 2z (a + a ) cos Ltl c Ltl h 2z (a a ) cos Ltl c Ltl 42 (6.27) (6.28)

An articial atom: the charge qubit In circuit QED system the role of the atom is taken by a superconducting circuit that includes a Josephson junction. This Josephson junction is nonlinear and its energy eigenvalues are therefore not equidistant, just like those of an atom. The cicuit we will discuss here is called charge qubit and consists of two superconducting electrodes separated by a thin insulating layer. In the superconducting regime, the electrons form Cooper pairs which are bosonic particles. Their Hamiltonian reads, 2 U (6.29) H = E j c c j + c c j (c c j 1) J(c c2 + c1 c ), j j 1 2 2 j j=1 where c j annihilates a Cooper pair in electrode j, [c j , c ] = jl , U describes the l Coulomb interaction between Cooper pairs averaged over their separations, J is the rate at which Cooper pairs tunnel through the insulating layer and E j is energy of Cooper pairs in electrode j. Instead of c j , the Cooper pairs can also be described in number and phase variables, c j = ei j n j with ei j = j =0 |n j n j + 1|. As can be seen from n i j , (ei j ) ] = |0 0|, the phase operator can not be Hermitian. Since devi [e ations from Hermitianity only appear for small particle numbers we can however ignore this complication and take to be Hermitian. One can furthermore show that particle number and phase behave like position and momentum, that is, [n j , l ] = i jl In terms of these variables, the Hamiltonian (6.29) reads,
2

(6.30)

H=

j=1

E jn j +

U n j (n j 1) J ei2 2

n1 n2 ei1 + ei1

n1 n2 ei2

(6.31)

We now introduce the new variables N = 1 (n1 + n2 ) and n = 2 (n1 n2 ). Since the 1 2 state of the Cooper pairs is close to a coherent state with high particle number, we can approximate N by a complex number, N = N , and assume n N. Conse 2 n2 N, and the Hamiltonian (6.31) simplies quently, we have, n1 n2 = N to, H = (E1 E2 )n +U n2 2NJ cos(1 2 ) (6.32) where we have dropped an irrelevant constant. Introducing the charging energy EC = U/4, the Josephson energy EJ = 2JN, the phase difference = 1 2 and ng = (E2 E1 )/(8EC ) as new variables, the Hamiltonian takes the standard form, H = 4EC (n ng )2 EJ cos() 43 (6.33)

Noting that 4EC n2 = 2(CQ g ) , we nd that EC = 2(CJe+Cg ) , where e is the elemen J +C tary charge (Note that the gate capacitance Cg and the Josephson capacitance CJ are in parallel for one of the electrodes.). Furthermore ng can be related to a volt2e age Vg applied at the gate of the junction by Vg = Cg ng . This gate voltage can be employed to tune the operating point of the junction. In a basis formed by eigenstates of n (n|n = n|n ), the Hamitonian (6.33) reads EJ (6.34) H = 4EC (n ng )2 |n n| (|n n + 1| + |n + 1 n|) 2 and one sees that for EC EJ the spectrum becomes deviates strongest form an 1 equidistant one for ng = 2 . Charge qubit and circuit cavity If the charge qubit discussed above is inserted into a circuit cavity, not only a constant gate voltage will apply to the Josephson junction but also the oscillating voltage U(z,t) of equation (6.28). Hence we have, ng = nDC + nAC g g where U0 =
h Ltl c

with

nAC = i g

Cg U0 (a a ), 2e

(6.35)

and we have assumed that the qubit is located at z = 0. The

Cg Hamitonian (6.33) generalises to H = 4EC [n nDC + i 2e U0 (a a )]2 EJ cos() g and includes a coupling to the photons. Expanding the quadratic term and apply-

ing a rotating wave approximation that is valid for


C C

2 Cg 2 2(Cg +CJ ) U0

h, we nd, [n
C2

g g g 2 nDC + i 2e U0 (a a )]2 [n nDC ]2 i 2e U0 (n nDC )(a a ) + 2(Cg +CJ ) U0 a a. g g g With these approximations the Hamiltonian of the charge qubit and the circuit cavity reads, H = HJJ + H phot + HI (6.36) g 2 with HJJ = 4EC (n nDC )2 EJ cos(), H phot = ( + 2(Cg +CJ ) U0 )a a and HI = h g g i2e Cg +CJ U0 (n nDC )(a a ). g g For nDC = 1 and EC |2e Cg +CJ U0 | the dynamics can be constraint to the subg 2 space spanned by states |0 and |1 of the charge qubit. Dening the states | = (|1 |0 )/ 2 and | = (|1 + |0 )/ 2 and the Pauli operators z = | | | | and = | |, the Hamiltonian (6.36) takes the form of a Jaynes-Cummings Hamiltonian, H = EJ z + h a a + (g+ a + g a ), with 2 g g 2 = + 2(Cg +CJ ) U0 and g = ie Cg +CJ U0 . Compared to atoms, the Josephson junc-

C2

C2

tions dipole moment is 104 times larger giving rise to much stronger couplings. 44

Chapter 7

45

7.1

Light Forces on Atoms

So far we have assumed that the emitters, Josephson junctions or atoms, are xed at a certain position in space while they interact with light elds. In practice however atoms can move. If a free atom is illuminated with a laser, it can absorb photons from the laser and re-emitt them either into the laser mode or the electromagnetic vacuum. In each such absorption or emission process the velocity of the atom changes by the recoil velocity vrec = hkL /m.

7.1.1

Concept of Doppler cooling

Photon absorption and re-emission processes together with the Doppler effect can be used to cool atoms. A moving atom will see the laser light at a Doppler shifted frequency L = L kL vL . Hence an atom that moves in the opposite direction of the laser photons, kL vL < 0, sees a higher frequency, L > L , than an atom at rest an vice versa. For atoms that are illuminated with a red detuned laser, L < 0 , this implies that atoms moving towards the laser source see laser light that is closer to resonance that atoms moving away from the source. The subsequent spontaneous emission occurs randomly in all directions and therefore has no effect when averaged of several absorption and emission events. The absorption in turn occurs predominantly when the atoms move towards the laser source where the atoms experiences a recoil of vrec = hkL /m in each ab sorption event. Since the recoil velocity is in the opposite direction to the atom velocity, the atom is on average slowed down.

7.1.2

Semiclassical theory of light forces


P2 h0 + z + h(R) + ei(Lt+(R)) + H.c. , 2m 2

The Hamitonian of a 2-level atom interacting with a laser eld reads, H = HA + HAL = (7.1)

where P is the momentum and m the mass of the atom. Here we treat the external degrees of freedom of the atom classically. That is we neglect its uncertainty in position R and momentum P. This is justied provided that, R L and kL v , (7.2)

where R is the position uncertainty, mv the momentum uncertainty, the spontaneous emission rate of the atom and kL and L the wavevector and wavelength of the laser. The rst condition says that the extension of the atoms wave function is small compared to L , whereas the second condition says that the atom travels 46

a distance much less than L between two spontaneous emission events. Both conditions are consistent with Heisenbergs uncertainty relation if, h ER , (7.3)

where the recoil energy reads ER = h2 kL /(2m). This is called the broad line 2 condition and says that the internal dynamics is much faster than the external motion. That is compared to the atom motion, the internal relaxation takes place instantaneously. Hence for studying the motion of atoms, one can assume that the internal degrees of freedom are at all times in the steady state of the master equation, i = [H, ] + (2 + + + ) (7.4) h 2 We identify the semiclassical force on an atom with the derivative of its potential energy with respect to its position mean value, i.e. F = HAL , r (7.5)

where r = R and p = P . From this denition we obtain, F = 2 (r) u(t)(r) + v(t) (r) h (7.6)

where (r) = 1 (r) r (r) is an intensity gradient and (r) = r (r) is a phase gradient. u(t) and v(t) are given by,

u(t) = Re

+ (t) ei[Lt+(r)] ,

v(t) = Im

+ (t) ei[Lt+(r)] ,

(7.7)

and are determined by the steady state solution of the equations, u = u + ( + )v 2 v = v ( + )u 2w 2 w = w + 2v , 2 which follow from equation (7.4). Here, = L 0 and w = turn we nd, dr = + = v t r dt r
1 2

(7.8) (7.9) (7.10) z . For in (7.11)

47

Atom at rest: For an atom at rest, v = 0 = 0, we nd the steady state solutions uss =
2

s , 2 1 + s

vss =

s , 4 1 + s

wss =

1 2(1 + s)

(7.12)

8 where s = 42 + 2 is the saturation parameter. Plugging u(t) = uss and v(t) = vss into equation (7.6), we obtain

F = FRP + FDP

with FRP = 2 vss h

and

FDP = 2 uss h

(7.13)

where FRP is called radiation pressure force and FDP dipole force. The radiation pressure force is dissipative since it can be expressed as FRP = p(e) , where p(e) is the occupation probability of the excited state and hence h p(e) the total rate of spontaneous emission. Since p(e) < 1/2, FRP is bounded. As an example for a plane wave laser with electric eld EL = E0 cos(Lt kL r) 22 one gets = 0, = kL and hence FRP = kL 2 2 +22 + 2 /4 . h The dipole force in contrast does not vanish in the limit 0. It can be h2 expressed via a potential, FDP = r VDP . For , we get VDP 2+ 2 /4 . The dipole force is unbounded as can be seen for and , where 2 / . As an example for a standing wave with electric eld E = VDP h L
L 0 2E0 cos(Lt) cos(kL r) one gets for kL ez , = 0 and FDP = 2 +82 cos2 (k

16 2 cos(k z) sin(kL z) h , 2 L z)+ /4 0

where 0 = d E0 . Friction force for moving atom We now consider a plane wave laser eld and an atom moving at velocity v, r = vt. There is no dipole force FDP = 0, but = kL v. We thus obtain the same steady state solutions as in equation (7.12) but with L replaced by L kL 22 v. the resulting radiation pressure force FRP = kL 2 h becomes 2 2 2
(kL v) +2 + /4

maximal for L kL v = 0 .

7.1.3

Standing wave Doppler cooling

We now consider Doppler cooling in a standing wave along the z-direction, EL = 2E0 cos(Lt) cos(kL z). As we have seen before, = 0 for this case and there is no radiation pressure force for an atom at rest. A moving atom in contrast can nonetheless experience a friction force that cools it. Since there are no analytical

48

solutions to eqs. (7.8) for this case we seek here an expansion in powers of, kL vz 2vz = 1. (7.14) = L To this end we dene, u 0 /2 0 u = v , s = 0 , B = /2 2 (7.15) w /2 0 2 and write equations (7.8) in the more compact form, u = Bu + s. Since the atoms move we have u = t u + vz z u vz z u, where we have taken into account that the internal dynamics is much faster than the motion of the atoms. We thus seek a solution to the equation, vz u = Bu + s (7.16) z in powers of . To this end we write u = u0 + u1 + . . . , where u0 is zeroth order in and u1 is linear order in . The contributions of zeroth order in to equation (7.17) read 0 = Bu0 + s and can be solved for u0 . The contributions of linear order in to equation (7.17) read vz z u0 = Bu1 and can be solved for u1 provided u0 has already been obtained. This procedure can be iterated further but we stop here at linear order. The resulting force on the atoms reads, F= hs 2 (1 s) 2s2 (2 + 2 /4) kL tan(kL z) 1 + kL tan(kL z)vz . 1+s (2 + 2 /4)(1 + s)2 (7.17)

Here the rst term corresponds to a trapping force where as the second term that is proportional to vz is a friction force for < 0 that results in cooling.

7.1.4

Limit of Doppler cooling

With the Doppler cooling concept described above, it is not possible to reach arbitrarily low temperatures. This is due to the fact that the spontaneously emitted photons on avergae carry a nite momentum of hkL . These spontaneous emission events thus give the atom a recoil of energy ER = 2mL . As these recoils go in random directions they correspond to a heating process that counteracts the cooling process. Doppler cooling will thus reach its limit when the associated heating rate ER p(e) becomes equal to the cooling rate |F v|. For low saturation s 1, one obtains from this condition that Doppler cooling can reach a minimal kinetic energy and hence minimal temperature for the atoms of, m 2 h h v and T , (7.18) 2 4 2kB where kB is Boltzmanns constant. This is the so called Doppler limit. 49
h2 k2

7.1.5

Cooling beyond the Doppler limit: Sisyphus cooling

The mechanism of Sisyphus cooling makes use of the fact that light of different polarisation couples to different atomic transitions. The total angular momentum J of the valenz electron is composed of a angular momentum L and a spin S, J = Each absorption and emission of a photon changes the angular momentum L + S. eigenvalue l (L| = hl(l + 1)| ) by 1, l = l 1. Hence for the simplest case, 1 one has as ground states two states with mJ = 2 and as excited states four state 1 3 with mJ = 3 , 1 , 2 , 2 . 2 2 The atom-photon coupling is VI = e r E. By Calculating the matrix elements of the commutators [Lz , z] = 0 and [Lz , x iy] = 1 (x iy) in the basis formed by h states |l, m , where L|l, m = hl(l + 1)|l, m and Lz |l, m = hm|l, m , one nds the equations, l , m ||l, m (m m) = 0 z and l , m |x iy|l, m (m m 1) = 0. (7.19)

These restrict the possible transitions. Hence for light with polarisation along ez which couples to atoms with d ez , only transitions with m = m have a non vanishing dipole matrix element and hence a non-vanishing coupling. In turn for light with polaristaion along ex iey , which couples to atoms with d ex iey , only transitions with m = m 1 couple. For the level structure introduced above, this means that -polarisation cou1 ples to the transitions |g, 1 |e, 3 and |g, 2 |e, 1 , -polarisation cou2 2 2 ples to the transitions |g, 1 |e, 1 and |g, 1 |e, 1 and + -polarisation 2 2 2 2 1 couples to the transitions |g, 1 |e, 2 and |g, 1 |e, 3 . The couplings are 2 2 2 however not all equally strong but carry weights given by Clebsh-Gordan coefcients. For the level structure at hand, the spontaneous emission rate on the 3 transitions |e, 3 |g, 1 and |e, 2 |g, 1 is for example 3 times larger than 2 2 2 1 on the transitions |e, 1 |g, 1 and |e, 2 |g, 1 . 2 2 2 For making use of these properties for cooling atoms, we put these atoms in a standing wave formed by two counter-propagating laser elds, one with momentum kez and polarisation along ex and one with momentum kez and polarisation along ey . The resulting standing wave then has a polarisation that alternates be tween and + with a periode of /4, where = |k|/2. That is, if the polarisation is at z0 , it changes to + at z0 + /4, back to at z0 + /2, again to + at z0 + 3 /4 and so on. Let us now assume that the atom is in one of the ground states, |g, 1 or |g, 1 . 2 2 Since the transitions for different polarisations have different coupling strength, the two ground state experience a different dipole potential, too. For ||, || , the diploe potential reads VDP h2 (z)/ and is thus proportional to the square of the dipole moment of the respective transition. Hence for -light, the state 50

1 |g, 2 is lower in energy (deeper dipole potential) than the state |g, 1 and vice 2 versa. An atom in state |g, 1 at a location with -light can thus only do the cyclic 2 1 transition, |g, 1 |e, 3 |g, 1 , and will therefore stay in state |g, 2 . 2 2 2 If this atom travels to a location with + -light, it has to climb uphill. At the + location with + -light it can however also do the transition, |g, 1 |e, 1 2 2 1 |g, 2 , and be transferred to the state |g, 1 . In this process the emitted photon has a 2 higher frequency than the absorbed photon. Hence the atom needs to loose kinetic energy. It is therefore cooled. This process is called Sisyphus cooling due to its analogy with the sentence of Sisyphus in the Greek myth.

Limit of Sisyphus cooling Let us denote by U0 the absolute value of energy difference between states |g, 1 2 and |g, 1 in a location with or + polarisation. On average the atom will thus 2 loose an energy of U0 /2 in each cooling cycle. However only 1/6 of all photon absorption and re-emission events contribute to the cooling. Yet each spontaneously emitted photon will heat the atom as spontaneous emission goes in random directions. Each such event increases the atoms kinetic energy by roughly 1ER . Hence there is a minimal value for U0 that is needed in order for the cooling to work. Doing more precise numbers, one nds min[U0 ] > 82 ER . The cooling process 5 will then come to a halt once the kinetic energy of the atom becomes small compared to U0 as the atom is then unable to climb the potential hills. At this point the atoms kinetic energy is about 1ER which is well below the Doppler limit.

7.1.6

Optical lattices

We consider a 2-level atom in a standing wave laser eld. The Hamiltonian for the internal degrees of freedom reads, H = z + h(z)(+ + ) Since this is a h2 22 matrix, it can be diagonalised. Its eigenvalues read, E = h 2 + 42 (z). 2 For ||
(z) h( || + || ). The ground state thus 2 experiences an energy shift 2 (z). One can associate a trapping potential to this energy shift. This is the dipole potential, which we have derived here in the dressed state picture. For a standing wave we obtain a periodic potential, that is called an optical lat-

|| and < 0, one nds E

tice, V = 0 cos2 (kz z). If we apply standing waves in all three spatial directions h we generate an optical lattice in 3d. A particle in this lattice has the Hamiltonian, H = 2 h2 2 + h 0 cos2 (k ), 2m =x,y,z 2 =x,y,z 51 (7.20)

Single-particle eigenstates We rst turn to nd the single particle eigenstates of the Hamiltonian (7.20). To model an innitely large system in the Thermodynamic Limit, we assume periodic boundary conditions. Since the Hamiltonian is a sum of terms for each coordinate = x, y, z we can make a product ansatz for the wavefunction, n,p = n,px n,py n,pz , where h2 2 +V0 (1 + cos(kx x)) n,px = n,px n,px , 2m x2
2

(7.21)

V0 = 0 and Hn,p = (=x,y,z n,p )n,p . The differential equation (7.21) h is known as Mathieus equation and its solutions can be expressed in terms of Mathieu-functions C and S, n,px = CCC(a, b, px x) + CS S(a, b, px x) (7.22)

mV mV Here CC and CS are expansion coefcients, a = h2 k0 h2m2 n,px and b = h2 k0 . 2 2 x 2 kx x Notice that k are the components of the laser wave vectors, whereas n and p label the possible solutions. For periodic boundary conditions with n,px x + 2 = n,px (x) only certain kx values for px are possible. These form bands that are labelled by n. For ultra-cold atoms and V0 > 5ER the width of the lowest band (n = 0) is smaller than the gap to the next higher band (n = 1). Hence a collision of two ultra-cold atoms in the n = 0 band cannot scatter one atom into a higher band. Therefore the dynamics of ultra-cold atoms is conned to the lowest band and we can write p = 0,p .

7.1.7

Many-particle representation: 2nd quantisation

We now turn to nd a representation for the many-particle version of the Hamiltonian (7.20). This reads H = N H j , where each H j has the form as in equation j=1 (7.20) and the index j labels the particles. The Hilbert space of the system is then spanned by product states of the form, | = |1 , 2 , . . . , N , meaning that particle j is in state | j . Here we focus on bosons only so that possible states of the many-particle system must be fully symmetric with respect to permutations of particles. We write such states as a sum over all possible permutations of particles, 1 (7.23) S|1 , 2 , . . . , N = P|1 , 2 , . . . , N , N! P where the prefactor 1/ N! has been introduced since the sum P contains N! terms. If several particles are in the same state ( j = l = . . . ) these states are 52

however not normalised. Correctly normalised and fully symmetric states can be written, 1 |n , n , . . . = S|1 , 2 , . . . , N (7.24) n ! n ! . . . This notation should be read as: n particles are in the state | , n particles are in the state | etc. That is, there are n values of j with | j = | etc. Importantly, the states are fully characterised by the values n , n , . . . . We can now dene the action of an annihilation and a creation operator on the states |n , n , . . . , (7.25) a | . . . , n , . . . = n | . . . , n 1, . . . We now express all operators that act on single or multiple particles in terms of a and a . Single particle operators have the form,
N

T=

j=1

Tj

with

Tj =

T , | j
,

j |,

(7.26)

where T , = j |T j | j is independent of j as all particles are identical. One has, N | j j || . . . , n , . . . , n , . . . = n n + 1 1 | . . . , n 1, . . . , n + 1, . . . j=1 n Here the prefactor n appears because a state | is converted to a state | in n -terms of | . . . , n , . . . , n , . . . as given by equation (7.24). The two other prefactors n + 1 and 1 then restore the correct normalisation of the state. n Hence a single particle operator T can be expressed as T=

j=1

Tj = T , a a .
,

(7.27)

In a similar way, each two particle operator V such as an interaction potential, can be expressed as,
N

V = V j,l =
j=l

, , ,

V , a a a a .

(7.28)

For a different basis, | = | | , the operator a = | a creates particles in the state | . Choosing | = |r to be position eigenstates, we dene the so called eld operator to be,
(r) = a = |r a = (r)a r

(7.29)

where (r) is the wave-function of state | . Hence the many-particle Hamiltonian H = N H j can be written, j=1 H= d 3 r (r) h2 2 +V (r) (r) 2m r2 (7.30)

53

Bloch and Wannier basis The eld operator (r) can be expanded in any set of basis states according to equation (7.29). Two important sets are the Bloch states k as dened in equation
1 (7.22) and their Fourier transforms, Wannier states w(r r j ) = N k eikr j k (r),

(r) = k (r)ak = w(r r j )a j ,


k j

(7.31)

where a j =

1 N

k eikr j ak . In the Bloch basis, the Hamiltonian (7.30) reads, H = k a ak ,


k k

(7.32)

whereas in the Wannier basis it reads H a a j J j


j
2 2

< j,l>

(a al + H.c.). j

(7.33)

h with J j,l = d 3 r w (r r j ) 2m r2 +V (r) w(r rl ) and = J j, j and J = J< j,l> . The mode energies of the Bloch modes are connected to and J by, k = 2J cos |k| and can therefore obtained by calculating the centre and width of the n = 0 Bloch band. They can be estimated to read,

3ER

V0 ER

and

4 V0 J ER ER

3 4

V0 ER

(7.34)

The Hamiltonian (7.33) thus describes particles that move on a periodic lattice, where we have only kept tunnelling terms between neighbouring lattice sites and neglected the much smaller long range tunnelling. For V0 ER the Wannier functions decay strongly as |r r j | increases. One can thus also obtaim by approximating in the potential cos2 (k ) 1 (k )2 and modelling each lattice site as a harmonic oscillator. For the tunnelling matrix element this approximation if for obvious reasons not good. So far we have assumed that the atoms do not interact. They however do interact via a van der Waals potenital.

7.1.8

Interactions between ultra-cold atoms

Here we consider the collision between two atoms of the same mass m. The Hailtonian reads, p2 p2 1 H2b = + 2 +V (|r1 r2 |) (7.35) 2m 2m 54

By dening centre of mass, R = 1 (r1 + r2 ), P = p1 + p2 and relative coordinates 2 r = r1 r2 ), p = p1 p2 , one can decouple the relative motion from the centre of mass motion. For the collision properties, only the relative motion is of interest, which is described by the Hamiltonian, p2 H= +V (|r|) (7.36) 2 The scattering of the atoms can thus be modeled as the scattering of an atom with reduced mass = m/2 at a given potenial V . To nd the scattering properties we 2 k2 2 k2 therefore seek solutions of the eigenvalue equation H = h2 , where E = h2 is the energy of the incoming atom that is scattered. Since the relative momentum p can be decomposed into a radial and an angular part, p2 = p2 + r2 L2 , with r k,l angular momentum L, we can make the product ansatz, k,l,m (r) = kr Yl,m ( , ). Here, the Yl,m are spherical harmonics that fulll L2Yl,m = hl(l + 1)Yl,m . We thus need to solve, h2 d 2 h2 l(l + 1) h2 k 2 k,l (r) + +V (r) k,l (r) = 2 2 2 dr 2r 2 (7.37)

The angular momentum enters here as a centrifugal potential r2 . The van der Waals potential V (r) is given by V (r) = C6 r6 outside the atom and has a hard core, V at the atom radius. For large r the repulsive centrifugal potential thus dominates. The sum of centrifugal and van der Waals potentials builds a potential barrier that is much higher than the kinetic energy of ultra-cold atoms whenever l 1. Therefore ultra-cold atoms can only undergo s-wave scattering with l = 0. We thus can ignore the centrifugal term in equation (7.37). h2 k2 For r much larger than the range of V , r r0 , we have V (r) 2 and can
h d approximate equation (7.37) by 2 dr2 k,0 (r) =
2 2

h2 k2 2 k,0 (r).

The solution reads (7.38)

k,0 A sin(kr + 0 ) for


2 2

r0
2 2

h d k For r k1 on the other hand, we have 2 dr2 k,0 (r) h2 k,0 (r). This can be seen by expanding k,0 (r) = cn rn , where terms with n < 3 do not contribute n=3 as they would not be normalisable. For ultra-cold atoms, typical ks are however small enough such that r0 r k1 exist. In this range we can thus approximate h2 d 2 (7.37) by 2 dr2 k,0 (r) = 0 and nd the solution

k,0 C0 (1 r) for r0

k1

(7.39)

In the range both solutions, (7.39) and (7.38) apply. We thus nd, k,0 C0 (1 r) = A sin(kr + 0 ) A(sin 0 + kr cos 0 ) for r0 r k1 where we have used kr 55 1. (7.40)

The phase 0 is called the s-wave scattering phase as it determines the phase relation between the incoming wave and the outgoing wave. For r r0 , c.f. 1 sin(kr+0 ) ei0 eikr eikr eikr (7.38), (r) = 4 = 2i4 e2i0 kr kr , where kr is an outgoing kr and e kr an incoming spherical wave. For s-wave scattering, the effect of the scattering potential is thus fully characterised by 0 . From (7.40) we nd, cot 0 = /k Instead of 0 one typically uses the s-wave scattering length aS dened by, aS = lim tan 0 , k0 k (7.42) (7.41)
ikr

to characterise the interaction potential. Here we have, aS = 1 . If we derive the s-wave scattering phase 0 for the contact-potential, V (r) = 4 h2 aS (r), 2 (7.43)

we nd cot 0 = 1/(kaS ) which is identical to (7.41). For ultra-cold atoms the interactions can thus be accurately described by the -potential (7.43). Feshbach resonances The interactions between ultra-cold atoms can furthermore be modied by the use of Feshbach resonances that occur because the interaction between ultracold alkali atoms depends on the relative orientation of the spins of their valence electron. Hence, at an energy, where there are unbound scattering states 2 k2 2 k2 ( h2 > V (r)) for the -conguration, there can be a bound state ( h2 < V (r)) for the -conguration. Since the Hyperne interaction couples the - and the 2 k2 -conguration, the scattering length changes dramatically as h2 approaches the energy of a bound state of the -conguration and vice versa. This is a Feshbach resonance. The relative energies of - and -conguration can be tuned by applying a magnetic eld and atoms can be driven through the Feshbach resonance. When crossing the resonance, the scattering length typically changes sign. One can thus switch between repulsive, aS > 0, and attractive, aS < 0, interactions.

7.1.9

Mott insulator to superuid quantum phase transition

We now consider ultra-cold bosonic atoms in an optical lattice that interact via a repulsive contact-potential as in equation (7.43). This system can undergo a 56

quantum phase transition from a Mott insulator to a superuid phase. The Hamiltonian describing this system is composed of the noninteracting part, see equa1 tion (7.33) and the interaction term given by, HI = 2 d 3 r (r) (r )V (|r 2 r |)(r)(r ) = U j a a a j a j , where U = 4 h aS d 3 r|w(r r )|4
2 j j m

8 kaS ER

V0 ER

3 4

.The full Hamiltonian now reads,

H = a a j J j
j

< j,l>

(a al + H.c.) + j

U 2

aj aj a j a j ,
j

(7.44)

and is called the Bose-Hubbard Hamiltonian. We have already encountered a 2-site version of it when discussing Cooper pairs in a Josephson junction, see equation (6.29). The Bose-Hubbard model describes bosonic particles in a lattices potential with short range interactions. This system resembles electrons in a solid, where the lattice potential is created by the ionised atom bodies and the electrons interact via Coulomb interactions. In our case the particles are however not Fermions like electrons but Bosons. Nonetheless, the Bose-Hubbard model has become a paradigmatic many-particle Hamiltonian. In optical lattices, the lattice sites are however spaced much further apart than in a solid and the lattice depth can be very accurately controlled via the laser intensities. The implementation in optical lattices thus offers unprecedented experimental control and measurement access. For U J and an integer number of particles per lattice site, the system is in a Mott insulating phase. Here the Hamiltonian can be approximated by H j a a j + U j a a a j a j . We assume a system of N particles and N lattice sites, j j j 2 that is on average one particle per lattice site. Then the ground state is |GS = j |1 j . Since adding a second particle to a lattice site would require an additional energy U, hopping of particles between lattice sites is suppressed and adding a further particle to the system requires and energy U. The second property is called incompressibility and is the dening property of the Mott insulator. For U J, the Hamiltonian can be approximated by (7.33) and the ground state of a system with N particles and N lattice sites is given by, 1 |GS = 1 (a )N |vac . This state is called a superuid. Since a = N j a j , N! k=0 k=0 all particles are delocalised across the lattice and there is a phase coherence between the lattice sites. The system can be driven through the phase transition between Mott insulator and superuid by tuning the ratio V J 2 E0 R (7.45) = e U kaS that can be modied by tuning V0 = via the laser intensity or detuning. 57
2

7.1.10

Measuring the Mott insulator and superuid phases

To measure which phase the ultra-cold atoms are in, the lattice potential is switched off and the atoms fall down due to gravity. While falling the atomic cloud expands ballistically since collisions are very unlikely once there is no longer a lattice potential. After falling a certain distance, an absorption image of the atomic cloud is taken by illuminating it with another laser and recording the shadow it creates. In ballistic expansion the location of an atom is given by r = vt = hk t. Since the m number of particles is conserved, d 3 r n(r) = d 3 k n(k), where n(r) and n(k) are the particle densities in position and in momentum space, one obtains, n(r) = m ht
3

n(k).

(7.46)

Hence the observed position distribution n(r) of the particles is proportional to the momentum distribution n(k) when the particles are released from the lattice. In free space (with the lattice off) one has,
2

n(k) =

d r

d re

ik(rr )

(r)(r ) =

d qe

ikq

w(q)

eik(r j rl ) aj al
j,l

(7.47) with Wannier function w(q). The difference between Mott insulator and superuid enters in the correlations a al . For a superuid of N particles on M lattice sites, j a al = j
N M

and one oberves interference peaks for k (r j rl ) = 2m with m


N M j,l

integer. For a Mott insulator in turn, one has a al = j peaks appear.

and no interference

58

Chapter 8

59

8.1

Trapped Ions

Ions are charged atoms and can therefore be trapped by applying and electromagnetic potential. As they couple directly via their charge to the electromagnetic elds, they can be subject to much stronger electromagnetic forces than neutral atoms which couple due to their polarisation.

8.1.1

Trapping potential: Paul trap

The ions are trapped in a minimum of an electromagnetic potential. Close to this minimum, the potential can always be approximated by a harmonic one, i.e. V (x, y, z) = V0 (kx x2 + ky y2 + kz z2 ). Since the potential V should full Laplaces 2 equation V = 0, one nds that kx + ky + kz = 0 which implies that at least one k is negative and no trapping occurs in that direction. Therefore only time-dependent potentials are necessary to trap charged particles in all directions of space. A prominent trapping conguration is a Paul trap, for which, V (x, y, z) = V1 V0 (kx x2 + ky y2 + kz z2 ) + cos(r f t) (px x2 + py y2 ), 2 2 (8.1)

with 0 < kz = (kx + ky ) and px = py . Typical values for the voltages are V0 0 50V and V1 100 500V whereas the radio frequency is typically r f 100kHz100MHz. The motion of a classical particle in the potential V as given in (8.1) can be found exactly. For the z-direction, it is a harmonic oscillation described by the equation of motion, Z|e| V0 kz z (8.2) z= m where Z|e| is the charge of the ion and m its mass. For the x- and y-direction, in turn, the equations of motion are Mathieu equations, e.g. x= Z|e| V0 kx +V1 cos(r f t)px x, m (8.3)
4Z|e|V0 kx 2 mr f

for which the solutions are known. We dene the parameters, ax = qx =


2Z|e|V1 px 2 mr f

and 1

and assume that the voltages V0 and V1 are chosen such that ax

and qx

1. In this regime, the solution of (8.3) can be approximated by, x(t) A cos ax + q2 r f x t 2 2 1 r f qx cos t 2 2 , (8.4)

60

and similarly for the y direction. The motion in the xy-plane thus decomposes into
r a so called secular motion at frequency ax + 2x 2 f and a much faster micror motion at frequency 2 f . Since qx 1, the micro-motions amplitude is also a lot smaller than for the secular motion. Therefore the micro-motion can be ignored in most cases. With these approximations, the potential for trapped ions can be approximated by a harmonic potential and the motion of the ions can thus be quantised in the standard way, H = ha az + ht (a ax + a ay ) (8.5) z x y r so that where = Z|e| V0 kz and t = ax + 2x 2 f . One can arrange for t m the transverse motion is in the ground state for cold ions. We now focus on this regime, where the transverse motion can be ignored and we arrive at an effective 1d model (a = az ), H = ha a (8.6)

q2

q2

61

8.1.2

Manipulations with lasers

Trapped ions interact with laser light via the dipole coupling as in eq. (5.4). Thus the location of the ion enters into the interaction. The Hamiltonian reads, H = ha a + h A z + h ei(L kL qA ) + H.c. , 2 2 (8.7)

where is the frequency of the ions oscillation in the trap, A the atomic transition
h frequency of the ion, the Rabi frequency of the laser and qA = 2m (a + a ) the quantised position of the ion in the trap. The Lamb-Dicke parameter,

= kL

2 h = 2m L

h 2m

(8.8)

h quanties the ration of the ions zero point motion amplitude 2m and the laser wavelength L and is typically 0.01 1. The Hamiltonian (8.7) can thus be expanded in powers of . For lasers of low enough intensity, such that , we can apply a rotating wave approximation. For different laser detunings we thus obtain the following interactions: For A L = 0 we drive so called carrier excitations ( = 1), h

H = a a +

2 A z + 2 a a x , 1 2 2 2

(8.9)

for A L = we drive the so called 1. red sideband, H = a a + A z i a + i + a, 2 2 2 (8.10)

and for A L = in turn we drive the so called 1. blue sideband, H = a a + A z i a + i + a . 2 2 2 (8.11)

For A L = 2 the second red and blue sidebands can be driven as well. By choosing suitable laser frequencies one can thus generate a Jaynes-Cummings interaction in Hamiltonian (8.10), c.f. eq. (6.2), and an anti-Jaynes-Cummings in Hamiltonian (8.11). Driving the 1. red sideband can be used for cooling as we will describe in the next section.

8.1.3

Sideband cooling

A laser that drives the 1. red sideband annihilates one phonon of the ions oscillation to excite the ion, see term + a in the Hamiltonian (8.10). If the ions 62

spontaneous emission rate obeys , the ions excited state subsequently 2 decays via spontaneous emission to its ground state before a phonon can again be created. The cycle thus reduces the ions vibrational energy by one phonon and can therefore be employed for cooling the ion. Provided the sidebands can be well resolved, i.e. , the process can cool an ion to its motional ground state.

8.1.4

Trapping multiple ions

One can not only load one but also multiple, say N ions into a trap. Here the trapping Hamiltonian generalises to,
N

H=

j=1 2m

p2 j

+V

with V =

m 2 2 1 Z 2 e2 zj + , 2 j=1 j=l 80 |z j zl |

(8.12)

where the rst part of the potential is the harmonic trap with axes along the zdirection and the second part is the Coulomb interaction. For cold ions their motion can be approximated by expanding the potential to leading order around the (0) equilibrium positions z j of the ions. Since the equilibrium positions are determined by
V z j z =z(0) j j

= 0, the potential is to leading order quadratic in the devia(0)

tions form the equilibrium positions, q j = z j z j . We nd,


N

H=
Z with A j,l = j,l + j,l 4
2 e2 0

j=1 m 2

2m + 2 2
2
(0) (0) |z j zk |3

p2 j

A j,l q j ql ,
Z (1 j,l ) 4
2 e2 0 m 2

(8.13)
2 (0) (0) . |z j zl |3

j,l=1

N k=1,k= j

The

matrix A can be diagonalised. It has non-negative eigenvalues m and eigenvectors b(m) , m = 1, . . . , N. The Hamiltonian (8.12) can therefore be written as a sum of (m) decoupled harmonic oscillators for the normal modes given by Qm = N b j q j j=1 and Pm = N b j p j . Hence in terms of creation and annihilation operators, j=1 mm i Qm + Pm . 2 h mm m=1 (8.14) The energetically lowest normalmode is always the centre of mass mode Q0 = 1 N q j with frequency 0 = . When adding laser drives that are tuned to the N j=1 rst red sideband with A L = 0 = , we obtain to leading order in , H= hma am m with m = m , am = H = 0 a a0 + 0 A 2
N j=1 N j=1 N (m)

zj i

2N
63

a + a0 . j j 0

(8.15)

Hence one can create interactions between ions via the centre of mass mode Q0 .

8.1.5

Ion trap quantum computer

Classical computers represent information as bits, 0 and 1, that correspond e.g. two values of a voltage. Hence a classical N-bit processor can process a bit string, 010011011100010010. . . , of length N at a time. Quantum computers on the other hand represent information in two possible states of a two-level system, |0 and |1 , so called qubits. A quantum processor of N qubits can thus process a state,
2

| =

j1 , j2 ,..., jN =1

c j1 , j2 ,..., jN | j1 , j2 , . . . , jN ,

(8.16)

where the sum extends over 2N orthogonal states. This indicates that a quantum processor is able to process all 2N possible classical bit strings in parallel. Equation (8.16) also shows that one needs 2N coefcients c j1 , j2 ,..., jN , that is 2N complex numbers to fully specify the state | . It is therefore very demanding to simulate the dynamics of a large quantum system, i.e. with large N on a classical computer. Moreover if one adds 1 qubit to the quantum system, the classical computer has to double its power to still simulate it since 2N+1 = 2 2N . If a classical computer can simulate a N qubit quantum system, it can also simulate a N qubit quantum computer and hence do all calculations the quantum computer could do. If the quantum computer is however increased by 1 qubit, the classical computer should double in power to still be able to solve all problems the quantum computer can solve. One can thus expect that quantum computers are more efcient than classical computers. In fact there are two quantum algorithms which are known to require less computational steps than any classical algorithm. Grovers algorithm can search an unsorted database of N entries in on average N steps, where a classical algorithm needs N/2 steps on average. Shors algorithm can factorise a number x into a product of prime numbers in a number of steps that grows polynomially in the number of digits of x. Classical algorithms need a number of steps that grows exponentially in the number of digits of x. Since the factorisation into prime numbers is unique, it is used in many encryption schemes which are secure due to the huge time it takes to nd the factorisation for large enough numbers. It is known that universal computation (roughly speaking everything one can expect a computer to do) can be done if single qubit rotations |0 + |1 |0 + |1 , , , (8.17)

and controlled NOT respectively controlled phase gates, |0, 0 |0, 0 |0, 0 |0, 0 |0, 1 |0, 1 |0, 1 |0, 1 respectively |1, 0 |1, 1 |1, 0 |1, 0 |1, 1 |1, 0 |1, 1 |1, 1 64

(8.18)

can be implemented. In fact, as shown by Cirac & Zoller in 1995, single qubit rotations and controlled phase gates can be implemented with trapped ions. To see this, we choose a chain of N trapped ions and let two of their internal states, say |g and |e represent a qubit, |0 j |g j and |1 j |e j j. (8.19)

Furthermore we consider an auxiliary level in each ion, |e . The transition |g |e with transition frequency e is thereby driven by a laser with one polarisation, say + and the transition |g |e with transition frequency e is driven by a laser with a different polarisation, say . We thus have N qubits, that can couple to each other via the centre of mass mode as described in the Hamiltonian (8.15). Single qubit rotations can be implemented for these ions by driving a carrier excitation as describe by the Hamiltonian (8.9). A controlled phase gate between ions j and l can be implemented by applying the sequence of unitary transformations, (1) (2) (1) U j,l = U j Ul U j , (8.20) where Uj
(1)

(2) Ul

= exp[ (|e j g j |a |g j e j |a )] 2 = exp[(|el gl |a |gl el |a )].


(1)

Hence to implement the unitary transformation U j , one applies a + -laser with Rabi frequency j to the rst red sideband of ion j for a time t = N/( j ). This is described by Hamiltonian (8.10) with L = e . To in turn implement (2) the transformation Ul , one applies a -laser with Rabi frequency l to the rst red sideband of ion l for a time t = 2 N/(l ). This is described by Hamiltonian (8.10) with L = e . The action of the unitary U j,l can be seen as follows: |g j , gl , 0 |g j , gl , 0 |g j , gl , 0 |g j , gl , 0 |g j , el , 0 |g j , el , 0 |g j , el , 0 |g j , el , 0 (1) (2) (1) |e j , gl , 0 U j |g j , gl , 1 Ul |g j , gl , 1 U j |e j , gl , 0 |e j , el , 0 |g j , el , 1 |g j , el , 1 |e j , el , 0 (8.21) where e.g. |g j , gl , 0 denotes a state with both ions, j and l in their ground states and no phonons present, a|g j , gl , 0 = 0. We thus see that equation (8.21) describes a controlled phase gate.

65

You might also like